Traditional economic models tend to focus on the accumulation of capital and labor. However, modern structural economics, particularly the Product Space Theory developed by Hausmann and Hidalgo, suggests that growth is driven by the accumulation of productive capabilities. These capabilities, ranging from specific technical know-how to infrastructure and institutions, cannot be traded directly. Instead, they are “revealed” through the products a nation is able to export competitively.
Under this framework, an economy is viewed not as a collection of isolated industries, but as a network of related activities. Products are connected by the similarity of the capabilities required to produce them. For instance, a nation capable of exporting textiles likely possesses the requisite “DNA” to export footwear, whereas the jump to electronics would require acquiring an entirely new set of non-tradable inputs.
The primary aim of this research is to decode this “Economic DNA” using unsupervised machine learning techniques. By analyzing global trade patterns from 2021, the study seeks to identify the hidden rules governing industrial diversification. Rather than relying on standard correlation metrics, Association Rule Learning (Apriori Algorithm) is employed to treat nations as “baskets” of capabilities. This approach allows for the detection of non-linear dependencies between sectors, effectively mapping the probabilistic paths of industrial development.
We used bilateral trade data from the Harmonized System (HS92). To analyze real specialization rather than just volume, raw export numbers are not enough (large economies naturally export more of everything).
To address this, the Revealed Comparative Advantage (RCA) index, also known as the Balassa Index, was calculated for every country-product pair. The RCA metric normalizes export flows by comparing a country’s share of exports in a specific sector to the global share of that sector.
An RCA value greater than 1 indicates that a country exports a product more intensively than the world average, revealing a comparative advantage.
## [1] "Analysis Year: 2021"## [1] "Total Export Flows: 21606"## [1] "Significant Links (RCA > 1) for Association Rules: 4236"For the purpose of Market Basket Analysis, the continuous RCA data required transformation into a binary format. A strict filter was applied: only trade flows with RCA>1 were retained as valid “transactions.” This step effectively filters out noise, products exported merely due to re-exportation or insignificant marginal production are discarded.
From an initial dataset of 21606 export flows, this filtering process isolated 4236 significant links, representing the core “productive matrix” of the global economy in 2021. These significant links form the input for the association rule mining.
To uncover the structural relationships between products, the Apriori Algorithm was utilized. In this context, the algorithm treats countries as “transactions” and the products they specialize in RCA>1 as “items.” The goal is to find rules of the form \(X \Rightarrow Y\), implying that if a country possesses the capability to produce , it likely possesses the capability to produce .
The strength of these connections is evaluated using three key metrics:
## Apriori
##
## Parameter specification:
## confidence minval smax arem aval originalSupport maxtime support minlen
## 0.5 0.1 1 none FALSE TRUE 5 0.07 1
## maxlen target ext
## 10 rules TRUE
##
## Algorithmic control:
## filter tree heap memopt load sort verbose
## 0.1 TRUE TRUE FALSE TRUE 2 TRUE
##
## Absolute minimum support count: 16
##
## set item appearances ...[0 item(s)] done [0.00s].
## set transactions ...[97 item(s), 234 transaction(s)] done [0.00s].
## sorting and recoding items ... [93 item(s)] done [0.00s].
## creating transaction tree ... done [0.00s].
## checking subsets of size 1 2 3 4 5 6 done [0.00s].
## writing ... [4693 rule(s)] done [0.00s].
## creating S4 object ... done [0.00s].Before examining specific industrial connections, the overall topology of the global trade network was assessed. The transaction matrix, constructed from countries possessing a Revealed Comparative Advantage (\(RCA > 1\)), exhibits a density of approximately 0.186. This sparsity is a fundamental finding: it quantifies the extent of global specialization. It indicates that the average nation possesses a comparative advantage in less than 19% of the analyzed product categories, confirming that industrial capabilities are unevenly distributed across the globe.
An analysis of the “most frequent items” reveals that primary resources and agricultural goods serve as the most common entry points into global trade. Categories such as Fish (03), Salt/Stone (25), and Edible Fruit (08) appear most frequently in national export baskets. This distribution suggests that while basic resource extraction is a ubiquitous capability, complex manufacturing remains the domain of a smaller subset of specialized economies.
## [1] "Transaction Data Summary:"## transactions as itemMatrix in sparse format with
## 234 rows (elements/itemsets/transactions) and
## 97 columns (items) and a density of 0.1866244
##
## most frequent items:
## 03 25 08 22 07 (Other)
## 94 92 82 79 78 3811
##
## element (itemset/transaction) length distribution:
## sizes
## 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
## 1 7 8 7 12 7 6 8 9 7 5 9 6 12 6 6 10 10 4 6 6 8 2 8 4 6
## 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 49 50
## 9 3 6 2 3 4 2 3 1 2 2 1 1 1 1 2 2 3 1 3 1 1
##
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 1.0 9.0 17.0 18.1 26.0 50.0
##
## includes extended item information - examples:
## labels
## 1 01
## 2 02
## 3 03
##
## includes extended transaction information - examples:
## transactionID
## 1 ABW
## 2 AFG
## 3 AGOThe application of the Apriori algorithm successfully identified the hidden “laws” of industrial co-occurrence. By filtering for high-confidence rules (\(Conf > 0.5\)) and sorting by Lift, distinct industrial clusters were revealed. The top rules exhibited Lift values exceeding 5.0, reaching as high as 6.21, which signifies extremely strong structural dependencies between sectors.
Two primary “Capability Clusters” were identified from the top 10 rules:
I. The Construction and Heavy Industry Cluster A robust network of dependencies was detected connecting raw material processing to finished industrial goods. Specifically, strong associations were found between:
Possession of advantages in these sectors is a near-perfect predictor for comparative advantages in Furniture (94) and Miscellaneous Articles of Base Metal (83). This logical progression, from the processing of raw inputs (steel, stone) to the manufacturing of complex assembled goods (furniture), demonstrates a clear value-chain integration. The high confidence levels (approx. 80-85%) imply that it is structurally difficult for an economy to export complex furniture competitively without first mastering the capabilities related to basic material transformation.
II. The Textile and Apparel Cluster A secondary, distinct cluster was observed within the light manufacturing sector. A strong rule links Articles of Apparel (62) and Footwear (64) to Headgear (65). This reflects a shared dependence on labor-intensive production methods and logistics networks common to the fashion and textile industries.
## [1] "Top 10 Export Rules (Strongest connections):"## lhs rhs support confidence coverage lift count
## [1] {68, 70, 73} => {83} 0.07264957 0.8500000 0.08547009 6.215625 17
## [2] {39, 73} => {83} 0.07264957 0.8095238 0.08974359 5.919643 17
## [3] {62, 64} => {65} 0.07692308 0.7826087 0.09829060 5.907433 18
## [4] {68, 73} => {83} 0.08547009 0.8000000 0.10683761 5.850000 20
## [5] {44, 73} => {94} 0.07692308 0.7826087 0.09829060 5.722826 18
## [6] {70, 73} => {83} 0.07692308 0.7826087 0.09829060 5.722826 18
## [7] {56, 73} => {83} 0.07264957 0.7727273 0.09401709 5.650568 17
## [8] {68, 73} => {94} 0.08119658 0.7600000 0.10683761 5.557500 19
## [9] {73, 94} => {83} 0.07692308 0.7500000 0.10256410 5.484375 18
## [10] {73, 83} => {94} 0.07692308 0.7500000 0.10256410 5.484375 18The objective of this study was to decode the “Economic DNA” of nations by analyzing the hidden structural dependencies within global trade data. Through the application of the Revealed Comparative Advantage (RCA) index and Association Rule Learning, the static trade flows of 2021 were transformed into a dynamic map of industrial capabilities.
The empirical results provide strong support for the path-dependence hypothesis proposed by Hausmann and Hidalgo. The discovery of high-lift rules linking heavy industry to furniture (\(Lift > 6.0\)) and apparel to footwear confirms that industrial diversification is not a random process. Instead, it is constrained by the underlying capabilities a nation already possesses. The high confidence levels observed in these rules suggest that economic development proceeds through “adjacent possible” steps, nations expand into products that share similar inputs, knowledge base, and infrastructure with their current export basket.
The low density of the transaction matrix (0.18) highlights a fundamental economic reality: specialization is the norm, not the exception. While basic capabilities (represented by frequent items like raw materials) are widespread, the distinct clusters identified by the Apriori algorithm reveal that advanced industrial capabilities are clustered in specific economic “neighborhoods.” This implies that the barrier to entry for complex industries is high, not merely due to capital costs, but due to the difficulty of acquiring the requisite network of non-tradable inputs.
This project shows that Market Basket Analysis, usually used in retail, works well for macroeconomics. Treating nations as “baskets” of capabilities allowed us to find structural links that simple correlations might miss. This approach offers a data-driven framework for strategic industrial policy, suggesting that nations should target sectors that are “mathematically close” to their existing advantages rather than attempting “leapfrog” development into unrelated fields.
hs92_country_product_year_2.csv dataset).