Machine learning to identify the most promising foreign markets

In today's data-driven era, Machine Learning is increasingly expanding its field of application thanks to its ability to manage and analyze vast amounts of information. Among the most well-known techniques is cluster analysis, which groups similar elements into “clusters,” within which each element shares common characteristics.
But how can cluster analysis support Market Selection?
As already discussed in the articles "Measuring Demand Potential: Big Data for Market Selection" and "Stages of Import Development: The Cases of Sweden, the United Arab Emirates, and Yemen", this technique can be used to assess the potential of target markets, identifying those that are expanding and have a higher likelihood of success for internationalization. Specifically, by focusing on the determinants of import growth, it is possible to analyze the life cycle of imports in a given country, thus anticipating the most favorable moments for market entry.

Identifying the stages of the import life cycle through cluster analysis

By taking the product life cycle as a reference, it is possible to outline a similar path for the development of its imports in a target country. Theoretically, this path is structured into three main phases: latency, growth, and saturation.

Based on the literature and the analysis of historical import data for various products in different countries, cluster analysis is expected to identify three groups of countries characterized as follows:

• Countries in the latency phase: characterized by a very low import-to-GDP ratio and import growth rates close to zero;
• Countries in the growth phase: have a still moderate import-to-GDP ratio but are accompanied by very high import growth rates;
• Countries in the saturation phase: exhibit a very high import-to-GDP ratio, with low import growth rates.

From a methodological perspective, the imports of individual products in different countries can be classified into each of these three phases through the implementation of cluster analysis, based on the following variables:

• Import share at constant prices relative to GDP in the year 2010;
• Import growth rate net of GDP growth, assessed over the last three years;
• Import growth rate net of GDP growth, assessed over the last ten years.

The clustering procedure was applied to the imports of consumer goods across all 150 countries covered by the ExportPlanning database. Once implemented, the algorithm highlights the presence of three groups of country-product combinations that exhibit characteristics similar to those theoretically outlined. Among these, the "Growth" cluster stands out, grouping countries with a still low import-to-GDP ratio but very high import growth rates (net of GDP growth).

A particularly significant insight comes from the fact that the median values of the three identified clusters align with the expected characteristics of products belonging to the three possible stages of import development. This result highlights the usefulness of cluster analysis: it provides a starting point for classifying country-product flows into different stages of import development and allows for the identification of critical thresholds for the variables used as input. Knowing these thresholds also enables an ex post validation of the classification of different flows into various clusters, allowing a degree of certainty to be assigned to each classification.

Naturally, the most interesting cluster is the one related to country-product flows in the “growth” phase. This cluster includes as many as 23,000 cases. Of these, 70%, approximately 15,000, were confirmed as certain based on a subsequent verification.

Which markets have the most products in the growth phase?

From the analysis results, it is interesting to note that the countries with the highest number of products whose imports are in the growth phase are Vietnam, Azerbaijan, the Philippines, Chile, and India. These five countries share multiple economic and institutional factors, including:

• Implementation of reforms aimed at increasing openness to foreign trade
• A per capita GDP in 2019 ranging between $2,000 and $5,000, or close to $15,000 in the case of Chile. This suggests the hypothesis of two per capita income thresholds beyond which significant growth in consumer goods imports can begin. The first threshold involves relatively lower-value products, while the second includes higher value-added products
• Demographic factors supporting national growth and significant economic growth rates

The following table presents the number of growing cases in the five considered countries, categorized by macrosector.

Number of products classified according to the Harmonized System (HS) imported by sector in the following countries during the growth phase of imports

The highest number of "growing" import cases concerns personal products imported by Vietnam. There are as many as 190 cases of personal product flows whose imports into Vietnam are "certainly" in the "growth" phase.

Conclusions

The use of machine learning techniques to analyze large volumes of data is particularly useful in identifying the stages of the import life cycle. Specifically, the growth phase helps pinpoint emerging countries for the import of certain consumer goods, where there is significant demand for these products and where adequate infrastructure is already in place to support trade and distribution. Cluster analysis thus serves as an initial step in classifying countries according to the stages of import development for a product, enabling the identification of rapidly changing and potentially promising markets for export.