Impact of the COVID-19 Pandemic on Export Survival from Latin American Countries

Abstract

This study analyzes the impact of mobility, as a proxy for social distancing measures, on exports to the United States of America (USA). A mobility index based on Google mobility indicators was constructed using Principal Component Analysis (PCA), and an Accelerated Failure Time (AFT) model was fitted to the data on export survival from a group of Latin American countries (LATAM). Higher mobility levels are associated with an acceleration of the risk of interruption of exports. On average, LATAM shows higher export survival levels compared to other countries. Higher innovation and market concentration favored export survival, while higher levels of Real Effective Exchange Rate (REER) are associated with a lower probability of survival. Differences in survival were found between export sectors with regard to machinery and transportation equipment.

1. Introduction

The COVID-19 pandemic poses a dilemma for many countries. Those that have prioritized health protection through public policies and implemented strict social distancing have experienced a significant reduction in their economic activities. By contrast, countries with loose sanitary measures to minimize disrupting economic growth from the pre-COVID-19 trajectory are recording highly negative results in terms of public health. Countries seeking a balance between both objectives would attain less severe results but might demand more resources [1,2,3].

It is important to consider that the measures to contain the spread of COVID-19 during the first months of the pandemic in LATAM, such as social distancing, mobility restrictions, temporary closure of activities, and reduction of working hours in non-essential activities, were not applied to the same extent across LATAM countries because each country determined which activities were essential by economic sector, as well as any changes in the working hours and temporary closure of activities. In most LATAM countries, activities such as tourism, construction, real estate, transportation, financial services, and education were affected for being considered non-essential.

In general terms, unemployment in production reduces supply; demand leads to contraction because the confinement decreases the frequency of use or imposes too many restrictions on the local environment. On the other hand, the potential winners under a pandemic include healthcare services, food processing and retail, personal care, products of greater economic complexity such as information and communication technologies [4], online commerce, and agriculture since they are deemed essential for the population [5].

The main objective of this study was to analyze the effects of mobility, as a proxy for social distancing and closure of activities, on exports, by estimating export risk and survival. In this regard, we asked the following research question: Are there differences in the risk of interruption/failure and survival/duration of exports between countries with different mobility levels as a proxy for social distancing measures applied during the first months of the COVID-19 pandemic?

The study period covered exports from January 2015 to April 2020. We mainly focused on the impact of the COVID-19 pandemic on exports to the U.S. market. To this end, we developed a mobility index, using both six Google mobility indicators per country and a Principal Component Analysis (a dimensionality-reduction method). This allowed obtaining a mobility index considering three strata—low, medium, and high—using the univariate Dalenius-Hodges method. With these mobility levels, the aim is to contrast the null hypothesis of no differences in survival rates between products according to the mobility level of the exporting country in April 2020 versus the alternative of differences in export survival according to the mobility levels of the countries of origin. Additional variables of interest are the market concentration level of countries exporting goods to the United States of America, the maritime connectivity levels of the different countries, innovation levels, infrastructure, and technical training as factors that increase export competitiveness and survival. Additionally, we explored potential differences in export survival according to the type of product or productive sector. An accelerated failure time (FFT) model was estimated. The duration analysis first displays estimates of survival rates, cumulative failures, hazard rates, and export survival graphs. Afterward, the drivers of hazard rates estimated using an AFT model are outlined.

LATAM countries ship more than 50% of their exports to the United States of America, except for countries such as St Vincent and the Grenadines, Cuba, Belize, Bolivia, Paraguay, Trinidad and Tobago, Panama, and Argentina. Of the countries in the region, only Mexico and Canada have important participation with 27.13% of total imports from the United States of America, ranking 2nd and 3rd in importance, respectively.

The first section of this work includes a brief introduction to the effects of containment measures such as social distancing and mobility restrictions on non-essential activities during the first months of the pandemic in LATAM countries. We analyzed the criteria regarding essential activities and time of implementation of social distancing measures with mobility restriction in LATAM and the effects of such measures on exports to the United States of America. The period and area of study, the objective, research questions, variables, and the model used are also described. The second section presents a review of the literature on international trade studies that explore the permanence of exports through survival and duration models with Kaplan-Meier and proportional hazard models known as Cox and Accelerated Failure Time (AFT) models. This section also describes the main explanatory variables of survival, risk of failure, and duration of exports used in previous studies, as well as the main contribution of this study to the specialized literature. The third section describes the sources of the variables and the methodology used to construct the dependent variable (duration of exports) and explanatory variables (real effective exchange rate, Herfindahl-Hirschman market concentration index, maritime connectivity index, innovation, exports by economic sector in dichotomous form, and mobility index), based on the Google COVID-19 community mobility indicator. The AFT model is also described and the interpretation of the risk acceleration; the exports survival time ratio is explained by the variables considered. The fourth section outlines the results of the AFT model for February, March, and April. The survival and risk functions are also displayed graphically, as well as the predictive margins at 95% statistical confidence. The fifth section discusses the consistency of the key findings with previous studies, highlighting the contribution to the specialized literature on survival, risk, and duration of exports in the first months of the implementation of social distancing and mobility restrictions to contain the spread of COVID-19 in LATAM. Finally, the sixth section draws conclusions on the use of variables that explain the survival, risk, and duration of exports supported by a literature review; the methods used, such as Principal Component Analysis and Accelerated Failure Time Models with Log-normal distribution; the results from model estimates; and the interpretation of the estimators and the acceleration rate of the risk and survival time of exports according to their statistical significance. This section also describes the scope and limitations of the study.

2. Literature Review

We reviewed and analyzed the literature on export survival available in the main databases, including Scopus, Web of Science, and Clarivate, among others, to identify the variables used (dependent and independent), type of data, the econometric technique used, region or country, sector, and authors (

Table 1. Studies based on export survival analysis.

Independent Variables	Country or Region	Econometric Technique	Study
Innovation, productivity	ASEAN countries	Probit model with fixed effects	[6]
GDP, GDP per capita, geographic distance, prices, initial value of exports.	China	Kaplan–Meier survival function, discrete-time duration models	[7]
Destination of the product, Herfindahl-Hirschmann Index	Africa and Latin America	Probit regression	[8]
GDP per capita, geographic distance, landlocked country, common border, common language, tariff	China	Kaplan–Meier survival function, discrete-time duration models	[9]
Price, real exchange rate, distance, GDP, tariff regimes, and region	Developing countries, European Union	Cox hazard model	[10]
GDP, geographic distance, common language, common border,	Turkey	Kaplan–Meier survival function, discrete-time duration models	[11]
Geographic distance, GDP, common language	China	Weighted quantile regression for censored data	[12]
GDP per capita, bilateral distance, trade cost, real exchange rate, product price.	China	Kaplan–Meier survival function, discrete-time duration models	[13]
GDP per capita, distance, relative real exchange rate, access to the sea,	United States	Kaplan–Meier survival function, discrete-time duration models	[14]
Initial value, foreign-owned company, Balassa’s RCA, real exchange rate, free trade agreement.	Georgia	Discrete-time duration models	[15]
GDP, geographic distance, initial trade value.	OECD countries	Discrete-time duration models	[16]
GDP, geographic distance, landlocked country, industry tariff rate on final products,	China	Kaplan–Meier survival function, discrete-time duration models	[17]
Geographic distance, GDP, GDP per capita.	Norway	Kaplan–Meier survival function, Cox hazard model	[18]
Geographic distance, GDP.	Norway	Kaplan–Meier survival function, Cox hazard model	[19]
Distance to export markets, common language, common border, GDP, tariffs	Russia	Kaplan–Meier survival function, Cox proportional hazard model, discrete-time duration models	[20]
GDP per capita, Balassa’s RCA, market access, market penetration, distance, contiguity, language.	114 developing countries	Kaplan–Meier survival function, Cox hazard model, linear probability model with fixed effects	[21]
GDP per capita, population, number of products,	23 major countries in the global trade of agri-food products.	Kaplan–Meier survival function, discrete-time duration models	[22]
R&D expenditure, royalty expenditure, capital goods and equipment import expenditure, advertising and marketing expenditure, Herfindahl Index on sales, GDP.	India	Discrete-time duration models	[23]
GDP, common language, border, colonial link, sea, geographical distance, real exchange rate, entry regulations. export and import costs	96 countries	Discrete-time duration models	[24]
Balassa’s RCA, population, initial value, GDP, GDP per capita.	Laos	Kaplan–Meier survival function, discrete-time duration models	[25]
GDP, geographic distance, common language	The Netherlands	Cox proportional hazard model, discrete-time duration models	[26]
Geographic Distance, Multilateral Index, headcount, GDP.	Sweden	Discrete-time duration models	[27]
Initial export value, number of target markets, product diversification, market share (Herfindahl index), product share (Herfindahl index), GDP, geographic distance,	Spain	Discrete-time duration models	[28]
Degree of product processing, export experience, geographic distance	China	Kaplan–Meier survival function, Cox proportional hazard model	[29]
Export intensity, foreign-owned enterprises, type of industry.	China	Kaplan–Meier survival function, discrete-time duration models	[30]
GDP, population, tariffs, distance.	Sweden	OLS, probit, Fixed-Effect Variance Decomposition (FEVD) model applied within a Heckman framework, zero-inflated negative binomial model	[31]
Economic Complexity Index, GDP per capita, GDP, initial value, market and product diversification, geographic distance.	60 countries	Kaplan–Meier survival function, discrete-time duration models	[32]
GDP, geographic distance, access to the sea, common language, common border.	Spain	Kaplan–Meier survival function, discrete-time duration models	[33]
Agglomeration, Balassa’s RCA	Malawi, Mali, Senegal, and Tanzania	Discrete-time duration models	[34]
GDP, GDP per capita, export value at the beginning of the period (spell), access to the sea, real exchange rate,	China	Kaplan–Meier survival function, Cox proportional hazard model, discrete-time duration models	[35]
Ad valorem transportation cost, GDP, tariff rate, relative real exchange rate, agricultural goods	United States	Cox proportional hazard model, discrete-time duration models	[36]
Foreign-owned companies, Hirschmann-Herfindahl Index, high-tech industries	Hungary	Discrete-time duration models	[37]
Patents, population, GDP per capita, geographic distance, common language, common border, actual exchange rate, number of exporters, initial value, freight rate	105 countries	Cox proportional hazard model	[38]
GDP, GDP per capita, real effective exchange rate	European Union	Cox proportional hazard model	[39]
Geographic distance, common language, GDP, number of export products, real exchange rate	European Union	Kaplan–Meier survival function, discrete-time duration models	[40]
Foreign companies, total productivity factor, Herfindahl Index, market penetration	United Kingdom	Discrete-time duration models	[41]
Intensive and extensive import margins, sunk costs, fixed costs per period	46 countries	Kaplan–Meier survival function	[42]
Labor productivity, capital intensity, foreign enterprises.	Finland	Kaplan–Meier survival function, discrete-time duration models	[43]
Initial value, geographic distance, border, common language, colonial history, GDP, real exchange rate, whether the country is a neighbor	82 developing countries	Cox proportional hazard model	[44]
GDP, GDP per capita, distance, common border, real exchange rate, number of products	Germany	Kaplan–Meier survival function, Cox proportional hazard model	[45]
Herfindahl Index, Gini Index, number of products	Developing countries	Kaplan–Meier survival function, discrete-time duration models	[46]
Supplier production cost, search cost, relative trade costs, initial purchase size	United States	Cox proportional hazard model	[47]
GDP, weighted trade	United States	Kaplan–Meier survival function, Cox proportional hazard model	[48]
Ad valorem transportation cost, GDP, tariff rate, relative real exchange rate	United States	Kaplan–Meier survival function, Cox proportional hazard model	[49]

). The study did not aim to carry out a comprehensive search of empirical studies but focused on the works conducted over the past twenty years. The main results are presented below.

The types of data range from clustered to panel data, which were first processed with the early econometric techniques such as survival analysis, the Kaplan-Meier function belonging to non-parametric estimators, and Cox proportional hazard models classified within the semi-parametric duration models. Our review also included novel econometric strategies like duration models for discrete-time data such as logit, probit, extreme value, Weibull and Gamma distribution functions (parametric regressions), and state-of-the-art methods such as the weighted quantile regression for censored data.

The independent variables considered in regression analyses were macroeconomic variables such as GDP, GDP per capita, real exchange rate, and the Gini Index; microeconomic variables such as characteristics related to industry or companies; science and technology indicators such as patents; trade indicators such as the Balassa Index, tariffs, freight rate, transportation costs; variables of gravitational models such as geographic distance; variables related to economic geography such as Hirschmann-Herfindahl-type market and product concentration indices; country features such as a colonial link to another country, historical regime, and whether the country shares a common language with its trading partner; demographic indicators such as population size; and governance and institutional indicators.

According to the state of the art of export duration and the objectives of this study, we used variables such as the effective real exchange rate index, mobility index generated from Google mobility indicators by country, Herfindahl-Hirschman Market Concentration Index, maritime connectivity index, infrastructure innovation indicators, and technology readiness as pillars of economic competitiveness.

Duration models applied to international trade are a recent research topic. The pioneering works on trade duration are studies [47,48,49] showing that U.S. commercial relations are significantly short-lived, which is not in line with conventional international trade theories.

According to Nitsch [45] and, Baldwin and Krugman [50], exchange rates may affect trade relations as changes in the former can reduce the entry costs of exports. In this sense, products from countries with currency depreciation would be expected to show a lower rate of exit risk or interruption in exports to the United States, at least in the short term or at the beginning. As the authors of [44] point out, the depreciation of the exporting country’s exchange rate implies an increase in the importing country’s purchasing power in the exporter’s currency; this makes the exported products more appealing and increases the probability of export survival over time. The authors of [13] found that changes in bilateral exchange rates between China and destination countries are associated with a longer duration of exports from Chinese companies. The authors of [50] found that the depreciation of a country’s currency reduces its risk of exit of its products from the export market or interruptions in export flow. The authors of [36] observed that the effect of the exchange rate on the duration of imports from the United States of America is 50% higher than transportation costs based on the Cox proportional hazard model, while estimates with a probit model showed that both effects are similar. However, The authors of [10] argue that the exchange rate is irrelevant to determining trade duration. The study by the authors of [51] indicates a marginal effect of the exchange rate on the risk of agri-food exports from the European Union. In other studies conducted by the authors of [15,24,38], the exchange rate is not statistically significant.

The COVID-19 pandemic is an unprecedented event in the recent history of international trade. There is still no empirical literature describing the effects of the pandemic on exports at the global and regional levels or to the United States of America. For the above, the present study aimed to analyze the impact of social distancing and suspension of work activities in companies that produce goods for export. To this end, a mobility index is proposed as a proxy for these pandemic prevention and mitigation measures, aiming to estimate the impact of these measures on exports to the United States of America based on export data for 234 countries or territories from January 2015 to April 2020.

The authors of [23] used the Herfindahl Market Concentration Index to contrast differences in the survival of exports of Indian firms according to the industrial concentration level. However, they did not find statistically significant differences in sectoral specificities and concluded that the export risk of companies is essentially similar across different sectors. The authors of [37] analyzed product survival from a multi-product company approach, using the Hirschman-Herfindahl index to determine whether the domestic market concentration is a major source for paying fixed export costs. These authors found that the level of domestic market concentration was not statistically significant in export survival.

The degree of maritime connectivity of a country determines the transportation costs of its export products and thus its competitiveness in the international market. It measures the degree of connection of countries to global maritime transportation networks. This index is calculated by the United Nations Conference on Trade and Development (UNCTAD) based on five components: number of vessels, container transportation capacity, maximum vessel size, and number of services and companies deploying container vessels in a country’s ports. In a study addressing the costs of international trade in OECD countries, the authors of [52] used the maritime connectivity index in a gravitational model as a variable reflecting the quality of maritime transportation in OECD countries. These authors found that reducing trade transportation costs requires improving the functioning of seaports, infrastructure, and shipping quality in each country.

Chen and Wei-Chih [38] used the number of patents issued by the U.S. Patent Office as a proxy for innovative capacity to explain the duration of exports from 160 countries to the U.S.A.; they leveraged a Cox proportional hazard model. This study reported that export duration increases with higher innovation levels.

Fabling and Sanderson [53] explored the impact of infrastructure on the performance of export companies in New Zealand using a Cox proportional hazard model. The authors generated a control variable to contrast differences in the survival of exports before and after the construction of a port. They found no significant effects on exports from the start of operation of the port. The publications analyzed in the literature review do not intend to cover all the studies on duration models in the context of exports but only to describe their main contributions related to the variables of interest included in the specified model with the respective economic analysis. Another aspect to highlight is that there are no studies addressing export duration models for LATAM under a pandemic scenario. The theoretical-methodological contribution of Besedes and Prusa [48] has been adopted to construct a model for estimating the duration of trade relationships between the countries in the LATAM region and the United States of America and identifying whether these trade flows are permanent or transitory. The central questions identified in the works applying the export duration model approach are: What is the exact time that trade flows are maintained? Are these trade relations long-term or short-term? Under a pandemic scenario, duration models produce empirical evidence of the duration of trade relations between LATAM countries and the United States of America until they stop exporting to this economy due to the COVID-19 pandemic, an event considered a failure. In this sense, the intention is to measure the magnitude of the survival of exports from LATAM to the U.S.A. and the probability of failure occurring, i.e., when trade relations are interrupted.

3. Research Methodology

The index was constructed based on Google COVID-19 Community Mobility indicators. Anonymized location data from mobile devices were compiled to estimate the distance to specific geo-referenced locations such as retail stores, recreation sites, supermarkets and pharmacies, parks, transit stations, and workplaces; for residential areas, the length of stay is measured (

Table 2. Description of variables.

Variable	Description
Retail & recreation	Variation in mobility related to restaurants, cafes, shopping centers, theme parks, museums, libraries, and movie theaters.
Grocery & Pharmacy	Variation in mobility related to grocery markets, food warehouses, farmer’s markets, specialty food shops, drugstores, and pharmacies
Parks	Variation in mobility related to national parks, public beaches, marinas, dog parks, plazas, and public parks.
Transit stations	Variation in mobility related to public transportation hubs such as subway, bus, and tram stations.
Workplaces	Variation in mobility related to the workplace
Residential	Variation in mobility related to the place of residence.

Source: Own elaboration based on data from [54].

). These data are expressed in terms of the percent variation relative to a baseline day prior to the pandemic outbreak; the baseline day is the median over five weeks from 3 January to 6 February 2020.

The PCA was used to construct the index for the period February–April. The Kaiser-Meyer-Olkin measure of sampling adequacy was 0.86 (p = 0.01), showing that the PCA was adequate. The first principal component accounted for 80.63% of the total variance in the data.

Table 3. Total explained variance.

Component	Initial Eigenvalues			Sums of the Squared Loadings of the Extraction
	Total	% of Variance	% Cumulative	Total	% of Variance	% Accumulated
1	4.838	80.635	80.635	4.838	80.635	80.635
2	0.587	9.775	90.410
3	0.229	3.814	94.224
4	0.174	2.897	97.121
5	0.121	2.016	99.137
6	0.052	0.863	100.000

Extraction method: Principal Component Analysis.

shows the total explained variance.

The first component showed a characteristic root greater than 1, accounting for 80.63% of the variance. To estimate the index, the regression method was used as a criterion from which the standardized coefficients were obtained. These coefficients were used to generate the mobility index scores shown in

Table 4. Matrix of standardized coefficients.

Variable	CP 1
Retail & recreation	0.194
Grocery & pharmacy	0.186
Parks	0.151
Transit stations	0.194
Workplaces	0.192
Residential	−0.192

Source: Own elaboration based on the results of the Principal Component Analysis.

.

These coefficients were substituted in the principal component equation together with the standardized mobility values in each country for each period from February to April:

$$Mobilit{y}_{it}=U_j{X}_{it}+,\dots ,+U_k{X}_{kt}$$

(1)

where $U_{j }$ is the jth standardized coefficient that makes up the index (Table 4), and $X_{it}$ is the mobility variable in the ith country during the period $t$. The mobility index was stratified using the univariate Dalenius-Hodges method considering three strata: low, medium, and high.

Monthly export data for 234 countries or territories for January 2015–April 2020 were obtained from [55]. Mobility indicator data were obtained from [54] Google Mobility Indicators for April 2020. Data with annual frequency for the effective real exchange rate, Herfindahl-Hirschman Index, connectivity, and innovation index were obtained from the databases in [56,57,58] and are listed in

Table 5. Summary statistics.

Variable	Description	Mean	Standard Deviation
Dexport	Duration of exports that takes the value of 1 if $t_i$ is an export observation and 0 if $t_i$ is a right-censored observation	0.38	-
SITC	Product code of the Standard International Trade Classification	-	-
February mobility	Mobility index in February	−0.01	1.00
March mobility	Mobility index in March	−0.08	0.98
April mobility	Mobility index in April	−0.06	0.96
REER	Real Effective Exchange Rate	98.68	13.7
Concentration	Herfindahl-Hirschman Index	0.11	0.12
Connectivity	Maritime connectivity index	29.11	26.81
Innovation	Competitiveness 12th pillar Innovation	3.35	1.26
Food	Binary variable: food and live animals, 1; others, 0	0.12	-
Beverages	Binary variable: beverages and tobacco, 1; others, 0	0.01	-
Raw materials	Binary variable: Raw materials, 1; others, 0	0.09	-
Fuels	Binary variable: mineral fuels, lubricants, and related materials, 1; others, 0	0.02	-
Oils	Binary variable: Animal and vegetable fat, oils, and waxes, 1; others, 0	0.01	-
Chemicals	Binary variable: chemicals, 1; others, 0	0.13	-
Manufactured goods_ material	Binary variable: Manufactured goods per type, 1; others, 0	0.2	-
Manufactured goods _equipment	Binary variable: machinery and transport equipment, 1; others, 0	0.22	-
Various manufactured goods	Binary variable: various manufactured goods, 1; others, 0	0.14	-

Source: Own elaboration.

.

The inclusion of the mobility index in the Accelerated Failure Time model allowed for controlling the changes in the population mobility in each country during the first months of the COVID-19 pandemic relative to a “normal” period. This is because this index measures changes in the mobility of the population during the first months of the pandemic—February, March, and April 2020—relative to the median mobility under normal conditions (the previous five weeks, from 3 January to 6 February 2020). The inclusion of the mobility index as a proxy for social distancing measures, changes in working hours, and the temporary closure of economic activities made it possible to identify the effects of these measures on the survival, risk, and duration of exports between LATAM and the United States of America.

Accelerated Failure Time Model

The Log-normal Accelerated Failure Time (AFT) model assumes that the natural logarithm of time follows a normal distribution; survival functions and the hazard function can be expressed as:

$$S\left(t\right)=1-\mathsf{\Phi }\left(\frac{\log \left(t\right)-\mu }{\sigma }\right)$$

(2)

$$h\left(t\right)=\frac{\mathsf{\Phi }\left(\frac{\log \left(t\right)}{\sigma }\right)}{1-\delta \left(\frac{\log \left(t\right)}{\sigma }\right)}$$

(3)

where $\mathsf{\Phi }(\cdot )$ is the standard normal cumulative distribution function and $\delta (\cdot )$ is the probability density function at time t. Given the matrix of covariates $X={\left(X_1,X_2,,\dots \dots ,X_P\right)}^t$, the survival function is:

$$S\left(t\right)=S\left[t exp\left({\beta }^{t}X\right)\right]$$

(4)

where $t\in T$ is the survival time, ${\beta }^t$ is a vector of coefficients, and $exp=\left({\beta }^{t}X\right)$ is the acceleration factor of the survival function. According to [59], the log-likelihood function to estimate, for $n$ observed times, $t_1,t_2,\dots , t_n$, is given by:

$$logl\left(t;\beta ,\mu ,\sigma \right)={{\displaystyle \sum }}_{j=1}^n\{-{\delta }_i\log ({\delta }_i{t}_i+{\delta }_{i}log{f}_i\left(t_i\right)+\left(1-{\delta }_i\right)log{S}_i\left(W_i\right))\},$$

(5)

where $W_i=\left\{log{t}_i-\frac{\mu +{\beta }_{1i}{X}_{1i}+{\beta }_{2i}{X}_{2i}+\dots {\beta }_{pi}{X}_{pi}}{\sigma }\right\},X=\left\{X_{ji}\right\}$ is a vector of covariates. In our implementation on the duration of exports, $X$ encompasses mobility, connectivity, concentration, REER, and innovation values for the j-th country; δ is a variable that takes the value of 1 if $t_i$ is an export observation and 0 if $t_i$ is a right-censored observation due to the absence of exports; ${\beta }^t=\left({\beta }_1, {\beta }_2,\dots , {\beta }_p\right)$ are the parameters to estimate. The interpretation of the acceleration ratio $exp\left({\beta }^{t}X\right)$ is the survival time. Mobility index values for each period are shown in Appendix A.

4. Results

Table 6. Results of the mixed-effects log-normal AFT models.

Variable	Mobility Model in February		Mobility Model in March		Mobility Model in April
	Coefficient	Time Ratio	Coefficient	Time Ratio	Coefficient	Time Ratio
Latin American Countries	-	-	-	-	0.0411 ***	1.04
Mobility	0.02	-	−0.09 *	0.92	−0.0691 *	0.93
Connectivity	0.01 *	1.01	0.00 *	1.00	0.0032 *	1.00
Concentration	0.72 *	2.04	0.81 *	2.24	0.5929 *	1.81
REER	−0.01 *	0.99	−0.01 *	0.99	−0.0102 *	0.99
Innovation	0.15 *	1.16	0.18 *	1.20	0.1911 *	1.21
Food	−0.11	-	−0.11	-	−0.1052 ***	0.90
Beverages	0.07	-	0.07	-	0.0757	-
Raw materials	−0.50 *	0.61	−0.49 *	0.61	−0.4893 *	0.61
Fuels	−0.57 *	0.56	−0.50 *	0.61	−0.5073 *	0.60
Oils	−0.17	-	−0.17	-	−0.1709	-
Chemicals	−0.05	-	−0.05	-	−0.0529	-
Raw material for manufactured goods	0.00	-	0.00	-	−0.0018	-
Various manufactured goods	0.10	-	0.09	-	0.0986	-
Constant	3.53	34	3.76	43.10	3.7487	43.95
February: number of observations = 8475; Log likelihood = −32480.879; Wald Chi2(14) = 1289.91; Prob > Chi2 = 0.0000. March: number of observations = 9734; Log likelihood = −36354.423; Wald Chi2(14) = 1579.95; Prob > Chi2 = 0.0000. April: number of observations: 10,160; Log likelihood: −37971.123; Wald Chi2(13): 1556.82; Prob > chi2 = 0.0000.

(*) significant at 99% confidence; (***) significant at 90% confidence.

shows the results of global exports and exports from LATAM countries based on AFT regression models or accelerated failure time, assuming that export interruptions follow log-normal distributions. The predictors were interpreted using time ratios (TR), where the ratio denotes the export survival acceleration factor so that the time ratio will be $exp\left(\beta \right)$ for predictors expressed in ordinal variables and 1 − $exp\left(\beta \right)$ when there are differences between control groups in relation to the reference group.

The above results reveal that as of April 2020, products originating from LATAM countries had a 4% higher survival rate compared with products from other countries with 90 percent confidence. Considering the same period, increases in the mobility index were associated with a 7% lower survival of global exports with 99 percent confidence. Products from countries with higher maritime connectivity levels did not show a reduction in the probability of survival, as indicated by the 0.0% time ratio. This suggests that greater maritime connectivity reduces the risk of export failure with 99 percent confidence. Increases in the Hirschmann-Herfindahl index of market concentration stimulate export survival, as evidenced by the 0.81% time-to-failure ratio. A 1% rise in market concentration increases the probability of export survival by 0.81% (p = 0.01).

On the other hand, countries with bilateral real exchange rate appreciations have a slightly lower probability of survival (1%) with 99 percent confidence. The results of the variable technological preparation reveal that a 1% rise in the level of innovation and technical readiness increases the probability of export survival by 0.21%.

The mobility time ratio indicates that higher mobility levels are associated with a lower export survival; that is, exports from countries with greater mobility have a lower survival rate versus countries with low mobility (Figure 1).

On average, LATAM countries show higher export survival rates than other countries. Higher innovation and market concentration levels improve the duration of exports. Rises in the real exchange rate increase the risk of export interruption, affecting duration and permanence. Differences in survival were found between sectors with regard to machinery and transportation equipment. Figure 2 shows the survival, hazard, and curves with minimum, average, and maximum mobility index values.

Figure 3 and Figure 4 show the predictive margins of Mobility, Connectivity, REER, Innovation, and Concentration.

5. Discussion

Our results about the effect of innovation on exports are consistent with the findings reported by the authors of [13]. Market concentration significantly accounted for the duration of exports, as opposed to the results obtained by the authors of [10,11], who found no significant effect of market concentration. Exports from countries with higher levels of real exchange rate display an acceleration of the risk of interruptions and, therefore, less permanence or duration. This finding is consistent with those by the authors of [36,45,49,50], who reported that products from countries with currency depreciation have a lower risk of export interruption. Therefore, further studies are necessary to delve deeper into the importance of real exchange rate policies during the pandemic.

The results of the effect of maritime connectivity on the probability of export survival during the first months of the COVID-19 pandemic indicate that greater maritime connectivity reduces the risk of a fall in exports. This finding is consistent with [52], highlighting that the quality of maritime transportation reduces costs by making exports more efficient.

In general, the time ratio for February was not statistically significant. For March and April, it was significant at the 99% confidence level, with similar values for both months; this indicates an increased risk of failure or interruption of exports due to increases in the mobility index.

The effect of mobility on the duration of exports in the context of the COVID-19 pandemic had not been examined previously for LATAM. We found that the predicted marginal mean decreases with high-mobility levels (Figure 3). Exports from countries with higher mobility levels increase the risk of interruption and, therefore, the survival time.

This indicates that countries with fewer restrictions on mobility through lockdown measures differentiated by essential or priority activities recorded an increase in the risk rate of interruption of exports and, consequently, in the duration or survival of exports.

The survival rate of LATAM exports to the United States during the first months of the pandemic was higher than for the rest of the countries. Exports of products from countries with high mobility levels in March, such as Chile, Costa Rica, Bahamas, Jamaica, Brazil, Canada, Trinidad and Tobago, Belize, Antigua and Barbuda, Mexico, and Nicaragua—many of which are LATAM countriesshowed a higher probability of export survival in April 2020 relative to other LA countries with low mobility levels such as Ecuador, Peru, Bolivia, Honduras, Argentina, the Dominican Republic, Paraguay, Colombia, El Salvador, Guatemala, Venezuela, and Uruguay.

6. Conclusions

Exports from countries with high mobility levels showed a smaller permanence. Exports from countries with higher market concentration have a higher probability of survival. Higher innovation levels favored export survival. Countries with higher levels of REER have a lower probability of export survival. The differences in the permanence of exports between sectors indicate that the impact of the COVID-19 pandemic in each country depends on the export basket and mobility characteristics, actual exchange rate levels, innovation, and market concentration.

It is observed that, in general, products from countries with real exchange rate appreciations have a lower probability of surviving in the North American market. Exports of products from countries with higher market concentration levels have, on average, a greater probability of survival compared to those with lower market concentration levels. The time ratio for products from countries with different values of maritime connectivity indicates that they have the same probability of survival in the North American market for the study period. Products from countries with higher innovation levels and technological readiness have, on average, a higher probability of survival than those with lower innovation levels. Innovation and maritime connectivity in Latin American countries have a greater effect on the probability of export survival. Given that the impact of the COVID-19 pandemic on the economy and social interrelations is uncertain and can only be visualized in the long term, decision-makers need to use new information communication technologies, specifically Big Data analysis, to identify the best strategies to face the health and economic crises. In this sense, the results of the present work highlight that countries with higher mobility levels do not show higher export duration; on the contrary, it seems that higher mobility levels are associated with an accelerated risk of interruption of exports.

A pending topic for future studies is the need to conduct post-pandemic research to explore in detail the effects of social-distancing measures involving mobility restrictions on exports from Latin American countries.