The dynamics of exporting: to continue or exit exporting? evidence from Chile

(1)

University of Groningen Faculty economics and business Master International Economics & Business

The dynamics of exporting: to continue or exit exporting?

evidence from Chile

(2)

1

Abstract

This paper contributes to the literature on export dynamics by focussing on the differences between firms that continue exporting and those that exit exporting using firm-level data from Chilean manufacturing firms covering the years 2001-2007. Based on the results of the logistic regression that predicts the probability of export exit it appears, firms that exit exporting experienced a negative shock in productivity growth in the year prior to exit. Also, firms that exit exporting experienced a negative shock in their export market revenues in the year prior to exit. However, the data used contains a large amount of outliers and different ways of dealing with these outliers give different results, limiting the robustness of the results.

(3)

2

1 Introduction

Due to micro-economic studies that study exporting behaviour, we know that firms engaged in exporting have to overcome fixed export costs, where the fixed export costs consist out of the export market entry costs and the yearly fixed export costs. Heterogeneous firms models of international trade, as used by Bernard et al. (2007) show the importance of export market entry costs in explaining which firms engage in exporting. Export market entry costs are associated with the costs a firm has to make to be able to export, e.g. setting up a foreign distribution network, administrative burdens, the adjustments of product designs to local consumer tastes or regulations and information requirements, as information about foreign markets and consumer tastes. Export market entry costs may vary by location, as they may vary with the exchange rates, quality of institutions and cultural differences. A firm decides whether or not to export to a particular market based on the fixed export cost of that particular market and its ability to overcome those costs.

Most recent research on firm-level characteristics is focussed on the relationship between exporting and firm performance. Firms that export have other characteristics compared with firms solely active on the domestic market. From firm–level data it appears that exporters compared to non-exporters have superior performance characteristics. Previous research has found that firms that export are in general more productive, larger and pay higher wages compared with firms that don’t export (Bernard & Jensen 1999, Bernard et al. 2007, Wagner 2005). Mayer & Ottaviano (2007) state that exporting is only for the happy few, as only a small exclusive group of firms is productive enough to overcome the fixed costs of exporting.

(4)

3 Most previous research has focussed on exporting and firm performance, only recently research starts to focus on export dynamics. Export dynamics deals with the entry in and exit out of foreign markets. From these studies it becomes clear that a large part of the exporting firms stops exporting after just one year. So did Eaton et al. (2007) study firm-specific export patterns by using Colombian data, from which they find new exporters to be extremely small in terms of their overall contribution to export revenues, and most of them do not continue exporting in the following year. They explain this pattern by stating that new exporters frequently experiment with small scale transactions and are just sufficient enough to cover the costs of testing the waters, which are likely to be less than the cost of long-term export contracts. One year exporters will therefore be smaller and less productive and a negative shock in their productivity and/or size will be enough to force them out of export markets.

Bernard and Wagner (1998) find that for German firms exporting today increases the probability that the firm will export tomorrow by 50% compared with firms that are not exporting today. However this advantages depreciates by two thirds each additional year. So after 1 year of exporting the probability of exit will increase each subsequent year. Wagner (2007) finds that firms that stop exporting were less productive in the previous year compared with firms that continue to export. Indicating the importance of the fixed export costs and a firm’s ability to overcome these costs. Ilmakunnas & Nurmi (2007) apply duration models to explain the duration until new plants start to export and the duration until they exit from their export markets using data on Finnish manufacturing. They find that larger, younger, and more capital intensive firms are more likely to engage in exporting and will survive in the export market for a longer period. They also find that foreign ownership increases the probability of export entry.

(5)

4 Table 1: Exporter dynamics in Chile

2001 2002 2003 2004 2005 2006 2007

Number of

Exporters 968 1,050 1,092 1,137 1,147 1119 1070

Number of Entrants 128 116 113 120 90 101

Number of Stoppers 124 90 128 99 91 103

Firm Entry Rate 12.19% 10.62% 9.94% 10.46% 8.04% 9.44% Firm Exit Rate 11.81% 8.24% 11.26% 8.63% 8.13% 9.63%

Source:ENIA, years 2001-2007

(6)

5

2 Theory and hypothesis

Theoretical literature about fixed export costs and exporting, models the role of foreign market entry and exit costs in the decision to enter or exit foreign markets by profit- maximizing firms (Baldwin 1988). This paper will follow the view that when the expected export revenues exceed the costs of exporting, a profit-maximizing firm will engage in exporting (Bernard & Wagner 1998, Bernard & Jensen 2001, Roberts & Tybout 1997, Melitz 2003). The costs of exporting are the fixed and variable costs associated with the exporting activities. A firm enters or continues being active in export markets only if the profits generated from exports are able to cover the costs associated with exporting, the fixed and variable export costs.

The fixed exporting costs exists out of:

1. The export market entry costs, e.g. setting up a foreign distribution network, identifying foreign market preference and the search costs of identifying transport companies;

2. The yearly fixed export costs, e.g. the yearly fee paid to foreign distributors, the rent paid for additional production capacity and the rent paid for new factories and/or machines needed when foreign market customer preferences differ from domestic market preferences.

The variable export costs are the costs associated with the extra input needed for the foreign sales, e.g. labour costs, materials etc.

A firms decision to enter an export market is based on the expected difference in profit when the firm exports compared to when it does not export (πit), which depend on the price of the products (Pit), the variable costs of producing that product (VCit), the profit maximizing quantity (Qit),, the fixed costs (Fit) and the foreign market entry costs (FENit). When for firm i in period t πit > FENit, it is profitable to engage in exporting, as the firms profit when engaged in exporting would be higher than the profit when the firm will be not engaged in exporting. Therefore a firm will engage in exporting when:

(7)

6 The decision to continue exporting in t differs from the initial export entry decision. When a firm is active in an export market in t – 1 it already paid the foreign market entry costs (FENit), so for the subsequent years it doesn’t face this sunk costs of foreign market entry anymore. Therefore the decision to continue exporting in t is based on Pit, VCit, Qit and Fit. A firm will continue exporting in year t when:

Pit · Qit ≥ VCit · Qit + Fit ( 2 )

A firms decision to exit an export market doesn’t only depend on the profit/loss it is making but also on the foreign market exit costs, indicated by FEXit. A firm will only exit the export market when the expected loss from exporting is larger than the foreign market exit costs. The foreign market exit costs consist out of the rent paid for non-transferable fixed assets, so assets which can only be used for that particular market, and other closure costs. For example contract contingencies with distributors or buyers and possible penalties that have to be paid for cutting agreements. Exiting an export market is equal to Qit = 0, so when a firm is not producing any products for its foreign market we assume that the firm has exited that particular foreign market. A firm exits its export market when:

πit ((Pit - VCit) · Qit)-Fit < - FEXit, ( 3 )

It is therefore possible that a firm will continue to export in period t while he is making a loss. In this case the foreign market exit costs will lead to an even higher loss compared to the decision to continue exporting. So a firm will continue exporting in t when:

(8)

7 2.1 Productivity and exporting

More productive firms have lower variable cost (VC), as capital and or labour are used in a more efficient way. Which makes it more likely that firms with a high level of productivity export. A higher level of productivity makes the additional costs of exporting easier to cover, as they will have a higher profit margin due to their lower variable costs.

((Pit - VCit↓)· Qit)-Fit, VCit↓ = πit ↑ ( 5 )

This view is also known as the self-selection effect. More productive firms are more likely to engage in exporting due to their higher profit margin. According to this view are firms which engage in exporting already outperforming non-exporters before they start exporting.

However other researchers argue that there is a positive correlation between exporting and productivity due to economies of scale and more intense competition, which makes firms that export more productive, also known as the learning-by-exporting effect. So do Girma et al. (2004) find that UK exporters are not only more productive before they start exporting, but they also find that exporting further increases firm productivity, indicating the learning-by-exporting effect.

Besides having lower VC and therefore higher profit margins, will high productive firms in general also be larger measured in output. According to the Solow model, nations/firms can only increase in gross domestic production, GDP, by an increase in capital per worker or an increase in productivity. Which makes it likely that more productive firms will be larger. Thereby we know from previous research that there exist a positive relation between firm size and exporting (Verwaal & Donkers 2001, Bernard & Jensen 2001). There are three different arguments indicating a positive relation between firms size and exporting. Firstly, from a resource-based view we expect that differences in resources are a possible explanation if a firm exports or not. Larger firms are likely to have the resource base necessary for exporting, where smaller firms may be restricted in their ability to serve foreign markets due to limited resources.

(9)

8 economies of scale argument can work in both ways, as economies of scale may also stimulate an export strategy for small firms. Small firms have more to gain from increasing their sales through exports. Larger firms will have less to gain, as they have already obtained relatively large outputs in their domestic market.

The last argument focusses on risk aversion. Larger firms will in general be less risk-averse compared with smaller firms, due to a larger size of operations and a greater spread of risks. Larger firms will in general be active in multiple markets with multiple products. A potential failure in a particular export market will therefore affect a larger firm less seriously than it will affect smaller firms.

From this we expect firms that quit exporting to be just productive and large enough to overcome the additional costs of exporting. A negative shock in their productivity and output or an increase in the fixed or variable export costs will therefore be enough to force them out of their export markets (Eaton et al. 2007). We therefore expect to find that firms that quit exporting experienced a negative shock in their productivity level and therefore size in the year before exit, t -1. Where productivity is measured as the production per employee.

Hypothesis 1: Firms that quit exporting in year t experienced a negative shock in productivity

growth in year t-1.

2.2 Innovation and exporting

(10)

9 ((Pit ↑ - VCit)· Qit)-Fit, Pit ↑ = πit ↑ ( 6 )

Process innovations may lead to improvements in a firm’s productivity, lowering the variable costs of producing. Process innovations will therefore improve a firm’s profit margin and its ability to overcome the additional costs of exporting.

((Pit - VCit↓)· Qit)-Fit, VCit↓ = πit ↑ ( 7 )

It is quite hard to measure a firms innovativeness, as firms are often quite secretive about their expenditure on innovation due to competitive reasons. We will measure a firms innovativeness by its number of patents, as a firm with a large number of patents is likely to be more innovative. We expect a firm that experienced a decline in their number of patent rights to lose some of its market power and therefore its ability to ask a price premium for their product. Thereby may a decline in the number of patent rights also have as a result that a firm is not allowed anymore to use a certain process which will have an effect on their variable costs. Both, a lower price and higher VC, will lead to a lower profit margin. So when a firm experienced a decline in the number of patents it is more likely to quit exporting.

Hypothesis 2: Firms that quit exporting in year t experienced a decline in their number of

patent rights in year t-1

2.3 Capital investments and exporting

(11)

10 A capital intensive firm is therefore also more likely to continue exporting, as when it quits exporting a part of its capital will become superfluous. This capital stock will not generate any income anymore while the firm still has to pay the fixed costs related to this asset, e.g. interest costs and maintenance costs.

πit ((Pit - VCit) · Qit)-Fit < - FEXit ↓ ( 8 )

When a firm makes an capital investment in t-1 it is less likely that he will leave its export market in year t. Because leaving its export market will lead to higher fixed costs per product, and therefore a lower profit per product.

((Pit - VCit)· Qit↓)-Fit, πit ↓ if Pit > VCit ( 9 )

Hypothesis 3: Firms that quit exporting in year t made less capital investments in year t-1 compared with firms that continue exporting.

2.4 Export revenues and exporting

The relationship between the growth in export revenues and continuing exporting is pretty straightforward. When a firm’s export revenues are increasing over the years a firm will be more likely to continue serving foreign markets as this market is growing. When a firms export revenues suddenly decrease, it is more likely that a firm quits exporting. As the extra revenues generated by their exports may be not sufficient anymore to cover the fixed exporting costs.

(12)

11 2.5 Promotion and exporting

The relation between promotion and advertisement expenditure may go in both ways. It may lead to more brand awareness abroad and a stronger brand which a firm can use to differentiate itself from their competitors. This may reduce price competition. Thereby will more brand awareness generate more sales, and therefore lower fixed costs per product. While at the other hand some researcher argue that promotion and advertisement don’t carry well over national boundaries, therefore only affecting domestic sales (Caves 1981; Salomon & Shaver 2005). Overall we expect to find a positive relation between promotion and the probability of exporting. So a firm that leaves it’s export markets will spend less on promotion.

((Pit - VCit)· Qit↑)-Fit, πit ↑ if Pit > VCit ( 10 )

Hypothesis 5: Firms that quit exporting in year t spent less on promotion in year t-1 compared with firms that continue exporting.

3 Data and Methodology

(13)

12 As we want to analyse the exit patterns of firms and the firm characteristics that have an influence on this pattern we will start by performing a cross sectional comparison between firms that quit exporting and firms that continue exporting. A firm is considered to be an export quitter in year t if this firm reported a positive amount of exports in year t-1 but not in year t. Continuing exporters in year t are firms that reported a positive amount of exports in year t-1 and t. This cross sectional comparison will make it clear on which characteristics firms that quit exporting differ from firms that continue exporting. Besides the cross-sectional comparison we will also test which firm characteristics increase the probability of exit out of export markets and by how much.

A firm that exports has two options, it can continue with exporting or it can exit its foreign market(s). We want to know the effect of productivity growth, becoming more innovative, capital investments, increasing export revenues and promotion on the probability of exit. To test the effect of these characteristics we cannot use a linear regression as the outcome of our model is not normally distributed but binomial, the outcome can only take the value zero and one. Where one indicates a firm that continues exporting and zero a firm that quits exporting. Due to this binary outcome we have to make use of generalized linear models which will convert the standard linear outcome (−∞,+∞), to the interval (0,1).

3.1 Generalized linear models

(14)

13 A disadvantage of making use of a logistic regression is that it requires much more data to achieve stable and meaningful results compared with a linear regressions. A linear regression in general generates stable results when 20 data points or more per predictor are used. While for a logistic regression 50 data points or more per predictor are necessary to achieve stable results. Because of the size of our dataset we don’t expect this to be a potential problem.

3.2 Interpretation coefficients

Besides the difference in the response variable between the linear regression and a logistic regression also the interpretation of the coefficients is different. The logistic regression results cannot be interpreted in the same way as a linear regression. By a linear regression the coefficient on x is β, which can be interpreted as a 1-unit increase in x increases the dependent variable by β. The coefficient in a logistic regression is measured in log-odds units. By a logistic regression β still represents the relationship between the independent variables and the dependent variable, however the difference with a linear regression is that the dependent variable is on the logit scale. The β estimates by a logistic regression represent the increase in the predicted log odds of the dependent variable that would be predicted by a 1 unit increase in the independent variable, holding all other independent variables constant. So will the β of productivity growth tell us the increase in the log-odds of the dependent variable, quit export, by a 1-unit increase in productivity-growth, holding all other independent variables constant.

The interpretation of logistic regression results can be simplified by converting the β coefficients from log-odds units into odds ratios. This conversion can be done by taking the exponential of the log odds which will give us the odds ratios. When the coefficients are in odds ratios, β represents the increases in odds of the binary dependent variable for every unit increase in the independent variable. So for a one unit increase in the independent variable, the odds of a firm quitting its exports increases by the odds ratio.

(15)

14 3.3 Conditional logistic regression

As we are dealing with unbalanced panel data we are normally supposed to use fixed or random effect to control for firm heterogeneity. However to estimate the β coefficients of our independent variables only the firms are used which quit their export market during the panel history. However, it is not possible for a firm to quit exporting each year, e.g. when a firm stops exporting in 2002 it can’t stop again in 2003 as it wasn’t exporting in 2002. So it is likely that the majority of firms quit their export market only once during the panel history. Therefore it is not possible to use a fixed effect logistic regression as one of the conditions for a fixed effect logistic regression is that the dependent variable must be measured on at least two occasions for each individual. Which means we have to use an ordinary conditional logistic regression (Magnac 2003).

3.4 The model

The conditional logistic regression used:

Yit =1 if firm i exits their export market in year t Yit = 0 otherwise

(16)

15 The independent variables used are calculated in the following way and consist out of the following variables from the ENIA survey:

1. Productivity growth is the percentage change in productivity compared with t-1, where productivity is measured by the production per employee. The production per employee is calculated by dividing the gross value of production (VBP) by the number of employees (EMPTOT).

2. Patent growth is the percentage change in a firms number of patents (PATIMP) compared with t-1.

3. Capital investment is measured by the total investment in land (CBNTER + CBUTER), buildings (CBNEDI + CBUEDI), machinery (CBNMAQ + CBUMAQ) and vehicles (CBNVEH + CBUVEH) minus the disinvestments in land (VBUTER), buildings (VBUEDI), machinery (VBUMAQ) and vehicles (VBUVEH), by dividing this value by total sales (INGTOT) we get the percentage of total sales which is used for investments.

4. The growth in export revenue is calculated by the percentage change in export earnings (INGEXP) compared with t-1.

(17)

16

4 Results

This section will present the cross-sectional comparison between continuing exporters and exporters that quit exporting, followed by the logistic regression results. The cross-sectional comparison will reveal the most notable differences between continuing exporters and exporters that quit exporting. Section 4.2 consists out of the logistic regression results and these results will indicate how much the probability of exit changes by an increase in one of the independent variables.

From the examination of the variables it appeared that the data used contains a considerable number of unusual data values, so called outliers. Which can be concluded from table 2, which contains the maximum and minimum values of each variable and the level of skewness. So does the growth in productivity has a maximum growth rate of 12,190% which is a practically impossible outcome and therefore most likely the result from a measurement error. The same yields for the growth in total sales and export revenue.

Table 2: Summary statistics variables

N Mean Max. Min. Skwns

Total revenues _37,307 _8,180,922 _{6,680,000,000} ₇₁₆ _37.99

Productivity _37,290 _96,987.46 _456,000,000 _246.83 _124.4

Productivity growth in % _28,324 _13.97 _12,190 _-99.12 _52.29

Productivity growth in prev. year in % _20,971 _12.5 _5,371 _-97.95 _25.7

Sales growth in % _28,340 _17.14 _99,204.65 _-99.91 _141.45

Sales growth in prev. year in % _20,983 _17.41 _99,204.65 _-99.91 _140.44 Investments in prev. year % total sales _28,340 _2.22 _1,002.39 _-959.58 _7.53 Promotion in prev. year % total sales _28,340 _0.62 _110.28 ₀ _12.79 Export revenues growth in prev. year

in % 4,517 276.06 349701.9 -100 42.59

Export-Total revenue ratio prev. year

in % 28,430 6.88 100 0 3.28

Number of employees _37,307 ₇₇ ₅₇₄₅ ₀ _8.44

Note: Total revenues and productivity are reported in current prices in thousands of Chilean Pesos, employees in the average number of employees, all other variables are measured in percentage growth, Source: ENIA, years 2001-2007

(18)

17 Rodriguez (2004), to detect mild and extreme outliers. Our first model will consist out of all observations, so without dropping any of the outliers. Model 2, will consist out of the observation that remain after dropping the extreme outliers, where extreme outliers are classified as observations which fall outside the range:

1. Larger than 75th percentile plus 3 times the interquartile range (Q3+3*IQR) 2. Smaller than the 25th percentile minus 3 times the interquartile range (Q1-3*IQR)

In model 3 all outliers will be dropped. Where an outlier is an observation which falls outside the range:

1. Larger than 75th percentile plus 1.5 times the interquartile range (Q3+1.5*IQR) 2. Smaller than the 25th percentile minus 1.5 times the interquartile range (Q1-1.5*IQR)

(19)

18 4.1 Cross-sectional comparison

The cross-sectional comparison will be a comparison between continuing and quitting exporters on a number of firm characteristics.

Table 3: Cross-sectional comparison

t test Mean N Mean Std.dev. P value Difference Total revenues Continuing exporters 5,322 28,500,000 124,000,000 0.3987 -5,500,000

Stopped exporters 635 34,000,000 312,000,000

Productivity Continuing exporters 5,320 178,652 2,267,388.00 0.0356 -296,803 Stopped exporters 634 475,455 7,937,313.00

Productivity growth Continuing exporters 5,320 17.53 211.62 0.1414 -15.82 Stopped exporters 634 33.35 490.02

Productivity growth in t-1

Continuing exporters 4,102 14.81 97.85 0.0119 -13.45 Stopped exporters 448 28.26 172.18

Sales growth Continuing exporters 5,322 12.84 49.57 0.0147 5.12

Stopped exporters 635 7.72 52.42

Sales growth in t-1 Continuing exporters 4,104 16.2 87.41 0.8062 1.05

Investments in t-1 Continuing exporters 5,322 4.43 24.05 0.0276 2.13

Promotion in t-1 Continuing exporters 5,322 1.13 3.88 0.0086 0.26

Stopped exporters 635 0.87 2.57 Export revenues growth in t-1 Continuing exporters 3,759 337.96 6,902.48 0.5449 250.29 Stopped exporters 279 87.67 566.07 Export-Total revenue ratio in t-1 Continuing exporters 5,322 35.29 34.46 0 24.02 Stopped exporters 635 11.27 21.06

Number of employees Continuing exporters 5,322 221 325 0 97

Stopped exporters 635 124 263

Note: author’s calculations Source: ENIA, years 2001-2007

(20)

19 Besides the difference in size, experience quitting exports a significantly slower growth in their sales in the year of exit compared with the firms that continued exporting. However the growth in sales for quitting exporters in the year of exit is still positive which is peculiar as they lose their foreign sales, meaning that their growth in domestic sales makes up for their loss in exports. Stopped exporters also invested significantly less money, spent less money on promotion and depend less on their export revenues in the year prior to exit.

Remarkable is the fact that stopped exporters productivity in the year prior to exit increased significantly faster as it did for continuing exports, which is the opposite of our expectations. We expected that more productive firms will be better able to overcome the fixed costs of exporting, making it unlikely to exit out of your export markets when your productivity increases. It appears that this increase in productivity is mainly caused by large fluctuations in the number of employees. However after correcting for outliers quitting exporters experienced a negative growth in productivity in year t-1, as can be seen from table 4 and 5. So this inexplicable result in table 3 is mainly caused by the outliers in our data.

Graph 1: Kernel density plot productivity growth

0 .5 1 1 .5 2 Pro d u ct ivi ty g ro w th i n t -1 0 5 10 x

(21)

20 This is also confirmed by graph 1, which compares the kernel density of productivity growth in t-1 for stopping and continuing exporters. We see from graph 1 that the majority of stopped and continuing exporters experienced approximately the same growth in productivity in the year prior to exit. However the average growth in productivity in t-1 for stopped exporters is significantly higher because there are more outliers with high growth rates for stopped exporters.

In accordance with our expectations is the fact that for quitting exporters their export revenues are less important as they are for continuing exporters. For continuing exporters 35% of their total sales is generated by exports while for quitting exporters this is only 11% in the year prior to exit. As we want to investigate the influence of the outliers in our data, table 4 contains the cross-sectional comparison results after dropping the extreme outliers, model 2. We conclude from table 4 that growth in productivity, investment in capital and promotional spending in the year prior to exit don’t differ significantly anymore from each other when quitting and continuing exporters are compared.

(22)

21 Table 4: Cross-sectional comparison after dropping all extreme outliers

t test Mean N Mean Std.dev. P value Difference Total revenues Continuing exporters 2,167 27,700,000 121,000,000 0.1594 14,400,000

Productivity Continuing exporters 2,167 131,566.60 1,529,598.00 0.5971 66,971 Stopped exporters 146 64,595.75 209,706.90

Productivity growth Continuing exporters 2,167 11.82 79.89 0.5754 3.74

Continuing exporters 2,167 4.91 26.94 0.1702 3.21

Stopped exporters 146 0.29 0.55 Export revenues growth in t-1 Continuing exporters 2,167 10.64 55.95 0 22.02 Stopped exporters 146 -11.38 72.87 Export-Total revenue ratio in t-1 Continuing exporters 2,167 35.57 34.28 0 24.79 Stopped exporters 146 10.78 17.9 Number of employees Continuing exporters 2,167 228 333 0.0004 99 Stopped exporters 146 129 192

Note: author's calculations Source: ENIA, years 2001-2007

(23)

22 Table 5: Cross-sectional comparison after dropping all outliers

t test Mean N Mean Std.dev. P value Difference Total revenues Continuing exporters 1,543 26,400,000 128,000,000 0.5369 8,400,000

Productivity Continuing exporters 1,543 143,293.40 1,808,860.00 0.7223 67,103 Stopped exporters 92 76,190.68 261,551.80

Productivity growth Continuing exporters 1,543 10.77 84.02 0.7051 3.34

Continuing exporters 1,543 4.68 21.23 0.0039 6.64

Stopped exporters 92 -1.96 24.39

Stopped exporters 92 0.21 0.33 Export revenues growth in t-1 Continuing exporters 1,543 3.89 45.29 0 23.51 Stopped exporters 92 -19.62 53.35 Export-Total revenue ratio in t-1 Continuing exporters 1,543 35 34.2 0 25.15 Stopped exporters 92 9.85 17.38 Number of employees Continuing exporters 1,543 216 328 0.0674 64 Stopped exporters 92 152 214

Note: author's calculations Source: ENIA, years 2001-2007

(24)

23 4.2 Logistic regression results

After comparing continuing exporters and quitting exporter on a number of characteristics we will now examine the effect of productivity growth, becoming more innovative, capital investments, growth in export revenues and promotional spending on the probability of exit. We will perform the logistic regression for all three models as also been done for the cross-sectional regression, which enables us to determine the effect of the outliers on our regression results. Thereby will we use robust standard errors in our logistic regression, as we expect to suffer from heteroscedasticity due to the high amount of outliers.

Table 6: Logistic regression results

Dependent variable: Quit export Model 1 Model 2 Model 3

Productivity growth previous year 0.062 -0.174 -1.133 **

Growth in patents -0.002 0.099 0.678

Capital investments -1.647 ** -3.844 1.63

Export revenue growth -0.003 -0.793 *** -1.164 ***

Promotion -1.345 -7.91 -9.215 Constant -2.592 *** -2.627 *** -2.903 *** N 3517 2313 1635 LR chi2 7.16 24.46 29.08 Prob > chi2 0.209 0.0002 0 Pseudo R2 0.0042 0.0225 0.0411 Notes: * p < 0.1; ** p < 0.05; *** p < 0.01

(25)

24 capital investment we find an odds ratio of 0.193. Which means that an extra 1% of total sales invested in capital in t-1 will lead to a 419% higher change that a firm will continue exporting.

All other variables in model 1 are insignificant. From the cross-sectional comparison which included all observations we already know that quitting exporters experienced a faster growth in productivity in the year prior to exit when compared with continuing exporters. This also appears from the logistic regression results, as model 1 shows a positive relation between growth in productivity and quitting exporting, however this relationship is not significant.

After dropping the extreme outliers from our regression we see from model 2 in table 6 that the amount of capital investments in the year prior to exit lost its significance, while growth in export revenue in the year prior to exit now significantly lowers the probability of exit. By converting the coefficient of export revenue growth to its odds ratio we find an odds ratio of 0.45, indicating that a growth in export revenue in t-1 of 1% increases the probability of still being an exporter in t by 121%.

After dropping all outliers, so each observations which falls outside the Q3+/-1.5*IQR range, we get the results presented under model 3. Now, growth in productivity in the year prior to exit also has a significant influence on the probability of exit. By converting the coefficient of productivity growth to its odds ratio, we find an odds ratio of 0.32. Which tells us that a growth in productivity of 1% in t-1 increases the probability of still being an exporter in t by 210%. Also the growth in export revenue in t-1 still significantly lowers the probability of exit. According to model 3 an increase in export revenues in t-1 of 1% increases the probability of still being an exporter in t by 220%.

(26)

25 4.3 Results after controlling for sector differences

To control for the fact that we are observing firms from different sectors in the manufacturing industry, the three regressions are also performed after including control variables for each sector. By including these control variables for each sector we correct for the possibility that the majority of export stoppers are active in a sector which is in general more productive compared with other sectors. The sectors are categorised by the 2 digit ISIC codes (Rev 3).

When we perform the logistic regression for the three different models again after controlling for different sectors we get the results presented in table 7. We can conclude from table 7 that the influence of our 5 independent variables, the effect of productivity growth, becoming more innovative, capital investments, growth in export revenues and promotional spending on the probability of exit doesn’t differ from the regression results without including controls for the different sectors.

(27)

26

Table 7: Logistic regression results

Model 1, incl. controls Model 2, incl. controls Model 3, incl. controls Dependent variable: Quit export

Productivity growth previous year 0.089 -0.285 -1.357 **

Growth in patents -0.002 0.063 0.009

Capital investments -1.44 * -2.853 5.007

Export revenue growth -0.003 -0.713 *** -1.058 ***

Promotion -444 -23.456 -32.225

Constant -3.014 *** -2.977 *** -3.432 ***

Industry controls

17 - Manufacture of textiles 0.691 ** 0.757 ** 0.949 **

18 - Man. of wearing apparel 1.007 *** 0.886 * 1.469 **

19 - Tanning and dressing of leather 0.446 0.461 0.394

20 - Man. of wood and of products of wood and cork 0.134 -0.099 0.100

21 - Man. of paper and paper products 0.363 0.068 0.678

22 - Publishing, printing and reproduction of recorded media 0.788 * 0.854 1.136 *

24 - Man. of chemicals and chemical products 0.302 0.371 0.558

25 - Man. of rubber and plastics products 0.648 ** 0.590 * 0.833 *

26 - Man. of other non-metallic mineral products 0.269 0.505 -0.027

27 - Man. of basic metals -0.205 -0.781 -0.062

28 - Man. of fabricated metal products 1.055 *** 1.072 *** 1.230 ***

29 - Man. of machinery and equipment n.e.c. 0.851 *** 0.804 ** 0.886 *

30 - Man. of office, accounting and computing machinery

31 - Man. of electrical machinery and apparatus n.e.c. 0.565 0.867 * 0.362

32 - Man. of radio, television and communication equipment

33 - Man. of medical, precision and optical instruments 0.642 0.428 0.414

34 - Man. of motor vehicles, trailers and semi-trailers 1.222 ** 1.579 ** 1.560 **

35 - Man. of other transport equipment 1.968 * 2.565 **

36 - Man. of furniture; man. n.e.c. 0.946 *** 0.635 0.545 N

LR chi2 3507 2305.000 1632

Prob > chi2 41.08 40.860 39.41

Pseudo R2 0.0081 0.009 0.0088

(28)

27

5 Discussion

Because we believe that the outliers are caused by measurement errors during the data collection, these outliers should therefore not be used for the calculation of our results. The discussion and conclusion of this paper will therefore be based on the results of model 3.

The results of model 3 reveal that a sudden decline in productivity raises the probability that a firm will exit out of it export markets, which is in accordance with the findings of previous research (Wagner 2007, Ilmakunnas & Nurmi 2007). A decline in productivity increases the costs per product, reducing the profit margin and the firm’s ability to cover the additional costs of exporting. From our logistic regression we are also able to conclude that a decline in sales to export markets increases the probability of exit, which is in accordance with our expectations. A decline in foreign sales will make it less beneficial to pay the yearly fixed costs of exporting, increasing the probability of exit.

We didn’t find any evidence that innovation and promotion influence the probability of exit. We expected to find continuing exporters to be more innovative compared with firms that quit exporting. Product innovators are expected to exploit their market power not only in their local market but also in export markets. Process innovations may lead to improvements in a firm’s productivity, lowering the variable costs of producing and its ability to overcome the additional costs of exporting. However, we didn’t find a significant influence of innovation on the probability of exit, which is not only contradictive with our expectations but also with the findings of Caldera (2010), who found a positive effect of innovation on the probability of export market entry. A possible explanation can be that innovation doesn’t matter in the decision to continue or exit exporting. Or it can be that our measure of innovativeness, the number of patents, is not a sufficient measure. More patents doesn’t automatically imply that a firms is also more innovative. R&D spending and the number of product and process innovations, which are used by Caldera (2010) may be better indicators of a firm’s level of innovation. Also for promotion there doesn’t seem to be a significant relation with the probability of exit exporting. Indicating that promotion is mainly focussed on the local market and less on foreign markets.

(29)

28 intensive firms to be more likely to exit exporting. A capital intensive firm will benefit from high sale levels as this reduces the fixed costs per product. Extra investments in t-1 will therefore lower the probability of exit. The results of model 1 show that an extra 1% of total sales invested in capital in t-1 will increase the probability that a firm will continue exporting, which is in accordance with the results found by Ilmakunnas & Nurmi (2007). However this relation is only significant when no correction is made for outliers. As already mentioned we expect the outliers to be caused by measurement and collection errors, limiting the robustness of this finding. Thereby is the relation found after correction for outliers between capital investments on the probability of exit positive instead of negative. So according to the results of model 3 an extra investment in capital increases the probability of exit, however not being significant. Which is the opposite of the results from model 1 and our expectation.

6 Conclusion

This paper tried to come up with new insights why some firms continue exporting while others choose to quit exporting by using Chilean firm-level data and performing a cross-sectional comparison and logistic regression. The cross-cross-sectional comparison gives insight on which characteristic firms that continue exporting differ from firms that stop exporting. It appears from the cross-sectional comparison that firms that stop exporting are smaller measured in number of employees, depend less on foreign markets and experienced a decline in productivity growth.

(30)

29 promotion both don’t have a significant influence on the probability of exit out of foreign markets.

(31)

30

References

Acuna E. & Rodriguez C. (2004) ‘A Meta analysis study of outlier detection methods in Classification’, Technical paper, Department of Mathematics, University of Puerto Rico at Mayaguez

Baldwin, R. (1988) ‘Hysteresis in import prices: the beachhead effect’, American Economic Review, Vol. 78 (4), 773-785

Bernard, A.B. & Wagner, J. (1998) ‘Export entry and exit by German firms’, NBER Working Paper No. 6538

Bernard, A.B. & Jensen, J.B (1999) ‘Exceptional exporter performance: cause, effect or both’, NBER Working Paper No. 6272

Bernard, A.B. & Jensen, J.B (2001) ‘Why Some Firms Export’, NBER Working Papers 8349, National Bureau of Economic Research

Bernard, A.B., Jensen, J.B., Redding, S.J. & Schott, P. (2007), ‘Firms in international trade’, Journal of Economic Perspectives, Vol.21, 105-130

Caldera, A. (2010) ‘Innovation and exporting: evidence from Spanish manufacturing firms’, Review of World Economics, Springer, vol. 146(4), 657-689

Cassiman, B. & Golovko, E. (2007) ‘Innovation and the Export-productivity Link’, CEPR Discussion Papers 6411

Castro, L. (2012) ‘Does licensing induce technological spillovers to domestic firms?’ University of Colorado at Boulder, Economics Department Paper

(32)

31 Caves, R.E. (1981) ‘Intra-industry trade and market structure in the industrial countries’ Oxford Economic Papers 33, 203–223

Eaton, J., Eslava, M., Kugler, M. & Tybout, J. (2007) ‘Export dynamics in Colombia: firm-level evidence’, Working Paper 13531

Girma, S., Greenway, D. & Kneller, R. (2004) ‘Does exporting increase productivity? A microeconometric analysis of matched firms’, Review of International Economics, 12(5), 855–866

Hanley, A. & Monreal-Pérez , J. (2011) ‘Are newly exporting firms more innovative? Findings from matched Spanish innovators’, Kiel Working Papers 1735

Ilmakunnas, P. & Nurmi, S. (2007) ‘Dynamics of export market entry and exit’, MPRA Paper 6060, University Library of Munich

Magnac, T. (2003) ‘Panel binary variables and sufficiency: generalizing conditional logit’, Research Unit Working Papers 0308, Laboratoire d'Economie Appliquee, INRA

Mayer, T. & Ottaviano, G. (2007) ‘The happy few: the internationalisation of European firms’, Blueprints, Bruegel, number 12.

Melitz, M.J. (2003) ‘The impact of trade on intra-industry reallocations and aggregate industry productivity’, Econometrica 71(6), 1695-1725

Pavcnik, N. (2002) ‘Trade liberalization, exit, and productivity improvements: evidence from Chilean plants‘, Review of Economic Studies 69 (1), 245-276

Roberts, M. & Tybout, J. (1997) ‘The decision to export in Columbia: an empirical model of entry with sunk costs’, American Economic Review, 87 (4), 545-564

(33)

32 Van Beveren, I. & Vandenbussche, H. (2009) ‘Product and process innovation and the

decision to export: firm-level evidence for Belgium’, Open Access publications from Katholieke Universiteit Leuven

Verwaal, E. & Donkers, B. (2001) ‘Firm size and export intensity’, ERIM Report Series

Wagner, J. (2005) ‘Exports and productivity: a survey of the evidence from firm level data’, International Trade 0504005, EconWPA