complexity

Overview

This project applies the economic complexity framework — originally developed by Hidalgo et al. (2007) and Hausmann & Hidalgo (2011) for cross-country product spaces — to Indian cities at the subnational level, using micro-data from the Periodic Labour Force Survey (PLFS) 2024.

The analysis produces five contributions:

1. Bangalore: Industry Space & Development Opportunities — Standard economic complexity mapping using co-location relatedness across Indian districts, identifying Bangalore’s industrial strengths and highest-potential development opportunities.

2. Five-City Comparison: Bangalore, Mumbai, Hyderabad, Chennai & Pune — Cross-city analysis of specialization patterns, diversification profiles, and development opportunities across India’s major technology and industrial hubs.

3. Delhi: Dual Formal–Informal Industry Space — A novel extension that builds two parallel industry spaces for formal and informal workers, revealing the dual structure of Delhi’s economy and identifying formalization pathways — informal specializations with the highest formal capability readiness.

4. District-Level Complexity, Growth & Caste (SHRUG 1990–2013) — A 23-year panel of 553 Indian districts using all four Economic Census waves. Two analyses: (a) does complexity predict subsequent growth? and (b) does caste stratification constrain the capability space? Uses four ECI algorithms (Eigenvalue, Fitness, GENEPY, ECI+) and supplements with PLFS 2024 caste-specific industry data.

5. IPL Player Space — Applying the same co-location and relatedness methods to the Indian Premier League, treating franchises as “countries” and players as “products.”

Part 1: Bangalore — Industry Space & SWOT

Methodology

For each 2-digit NIC industry in each Indian district, I compute:

• Revealed Comparative Advantage (RCA): Is industry \(i\) overrepresented in city \(c\) relative to all-India?

• Co-location Proximity (\(\varphi\)): How related are two industries, based on their tendency to co-locate across 744 Indian districts?

• Density (\(\omega\)): How many related industries does the city already have?

Industries are then classified into a SWOT framework based on RCA (above/below 1) and Density (above/below median).

Industry Space Network

The nodes represent 2-digit NIC industries; edges connect industries with co-location proximity \(\varphi > 0.25\). Node size is proportional to RCA; colour indicates SWOT category.

SWOT Classification & Development Opportunities

Each dot is an industry. The four quadrants represent Strengths (high RCA, high density), Weaknesses (high RCA, low density), Opportunities (low RCA, high density), and Threats (low RCA, low density). The bar chart ranks the top development opportunities by density.

Key Findings

Top Strengths (RCA > 1, high density):

Electronics & optical products emerges as the top opportunity — Bangalore has RCA = 0.81 (nearly a specialization) with the highest density (0.635) of any industry, indicating strong capability readiness. Pharmaceuticals and financial services follow.

Part 2: Five-City Comparison

Cities Analysed

The economic complexity framework is extended to five major Indian cities, all computed from a single all-India proximity matrix (744 districts \(\times\) 80 industries). Each city is identified as the largest urban employment district in its state:

City Complexity Profiles & Cross-City RCA

The SWOT breakdown (left) reveals distinct industrial characters. Mumbai has the broadest base (57 specializations, 32 strengths) while Pune and Bangalore have the most untapped opportunities. The heatmap (right) shows which industries each city specializes in — 11 industries are shared across all five cities.

SWOT Comparison: Five Panels

Each panel shows one city’s RCA-vs-density scatter. The quadrant structure identifies Strengths (top-right), Opportunities (bottom-right), Weaknesses (top-left), and Threats (bottom-left).

Industry Space Networks

The same force-directed layout is used across all five cities, enabling direct visual comparison of which parts of the industry space each city has activated.

Key Patterns

Unique specializations (RCA > 1 in only one city):

Top specialization by city:

Development Opportunities

The opportunity comparison highlights each city’s highest-density industries where RCA is currently below 1 — the most promising targets for industrial diversification.

Part 3: Delhi — Dual Formal–Informal Complexity

Motivation

Approximately 86% of Delhi’s workforce is informal. Standard economic complexity analysis treats all workers identically, obscuring the dual structure of developing-country cities. This analysis builds separate industry spaces for formal and informal workers.

Informality Classification

Following the hybrid NCEUS/ILO framework applied to PLFS 2024:

Dual Industry Space

Two parallel industry spaces, using a shared network layout for visual comparison. The formal space (left) is sparser and concentrated in services; the informal space (right) is denser with strong manufacturing clusters.

Formal vs. Informal RCA & Dual Classification

The scatter (left) plots each industry by informal RCA (x-axis) vs. formal RCA (y-axis) — the diagonal represents equal specialization. The bar chart (right) shows the four-way classification based on where RCA exceeds 1.

Formalization Pathways

The most policy-relevant output: among Delhi’s informal specializations, which have the highest formal density — i.e., the strongest surrounding formal capabilities that could support a transition?

The top formalization pathways are:

1. Domestic household employment — highest formal density (0.51) with strong informal RCA (6.0)
2. Travel agencies — already strong in both sectors, easiest transition
3. Real estate — growing formal ecosystem supports informal workers
4. Education — 56% informal share but established formal infrastructure
5. Health activities — similar dual structure with 56% informality

Part 4: District-Level Complexity, Growth & Caste (SHRUG 1990–2013)

Data

India’s Economic Census (EC) was conducted in 1990, 1998, 2005, and 2013, covering every non-agricultural establishment in the country. The SHRUG platform (Asher, Lunt, Matsuura & Novosad) harmonizes these four waves into 90 SHRIC industry codes at consistent PC11 district boundaries. I construct a balanced panel of 553 districts observed across all four waves — approximately 553 \(\times\) 90 \(\times\) 4 = 199,080 district-industry-wave observations.

Supplementary data: DMSP calibrated night lights (1992–2013) and Population Census 2011 (literacy, SC/ST population share, total workers) from SHRUG; PLFS 2024 unit-level data for caste-specific industry disaggregation.

Methodology

For each EC wave, I compute a district \(\times\) industry employment matrix and apply four complexity algorithms in parallel:

Cross-correlations (EC 2013): ECI–GENEPY \(r = 0.45\), ECI–ECI+ \(r = 0.18\), GENEPY–ECI+ \(r = 0.42\). The algorithms capture different facets of complexity.

Part 4a: Does Complexity Predict Growth?

ECI is computed per district per wave, then regressed on subsequent employment growth (log-difference between adjacent waves, annualized) and night-lights growth. Specifications include cross-sectional OLS with state FE and panel regressions with district FE.

Top-ECI districts in 2013 are metros — Mumbai, Delhi, Bangalore, Chennai, Pune — validating that the measure captures economic sophistication. The trajectory plot tracks how these districts evolved over the 23-year panel.

Part 4b: Caste Stratification and the Capability Space

The Economic Census records aggregate SC, ST, and OBC enterprise counts per district (not per industry). Cross-sectional regressions (EC 2013, state FE) show strong negative associations:

Districts with higher marginalized-caste enterprise shares have systematically lower economic complexity.

Panel regressions with district FE reveal a striking pattern: the SC coefficient flips positive within-district (\(+1.05\), \(p = 0.036\)), while ST and OBC remain negative. First-difference estimates confirm: \(\Delta\)SC = \(+2.17\) (\(p < 0.001\)). This suggests that as districts become more complex, SC enterprises enter expanding sectors — a compositional dynamic rather than a causal effect.

PLFS 2024 Supplement: Caste-Specific Industry Spaces

The EC limitation (aggregate caste data) is addressed using PLFS 2024, which has individual-level caste \(\times\) industry. Separate employment matrices are built for SC, ST, OBC, and General workers, and caste-specific RCA and ECI are computed.

General-caste workers specialize in professional services (real estate, legal/accounting, finance), while SC workers are overrepresented in construction, beverages, and public administration. The industry stratification chart shows the 30 most caste-segregated industries:

The General–SC ECI gap is modest at the aggregate level (mean 0.024), but the real stratification is in which industries each group occupies: General castes cluster in high-complexity professional services, while SC workers are concentrated in low-complexity manual sectors.

Part 5: IPL Player Space — Complexity Meets Cricket

Can the economic complexity framework map a cricket league? This extension applies the same co-location and relatedness methods to the Indian Premier League (IPL), treating franchises as “countries” and players as “products.” A bipartite Teams \(\times\) Players network built from ball-by-ball data across 17 seasons (2008–2024) reveals how 724 players connect 15 franchises through shared rosters, team-switching, and co-occurrence. The resulting Player Space — where edges link players who have been teammates, weighted by shared seasons — surfaces clusters of franchise loyalty, player mobility ecosystems, and recruitment signatures distinct to each team.

The Player Space Network

The hero visualisation: 149 of the most active IPL players, connected by co-occurrence on the same team. Node colour marks each player’s primary franchise; node size reflects career length. The dense core shows high-mobility veterans who have represented multiple teams, while the periphery reveals franchise-loyal clusters — the CSK (yellow) and MI (blue) groupings are particularly tight-knit.

Franchise Similarity & Loyalty

Which franchises share the most players? RCB and DC share the most (44), followed by DC–KKR (39). The loyalty profile shows how many players were one-club vs. journeymen — CSK has the smallest roster (103), consistent with their reputation for stability.

Player Journeys

A dot-strip chart tracking the top 20 most experienced IPL players across seasons. Some players are strikingly loyal — MS Dhoni (CSK, 16 seasons), Virat Kohli (RCB, 16 seasons), Rohit Sharma (MI, 16 seasons) — while others like Ajinkya Rahane (6 franchises) and Dinesh Karthik (6 franchises) are quintessential journeymen whose mobility stitches the league together.

Most Connected Players & Squad Utilisation

David Warner leads with 31 unique teammates and the highest weighted co-occurrence strength. The heatmap shows squad sizes per season — early years saw larger squads (RR: 36 in 2009), converging to 17–22 by recent seasons.

Data & Methods

• PLFS 2024 (Parts 1–3): Periodic Labour Force Survey, January–December 2024, NSO, Government of India. 415,549 persons; 164,046 employed. 2-digit NIC 2008 (80 industries), 744 districts.
• SHRUG Economic Census (Part 4): DevDataLab harmonized Economic Census 1990, 1998, 2005, 2013. 90 SHRIC industry codes, 553 balanced districts, PC11 boundaries. Supplemented by DMSP night lights and Population Census 2011.
• PLFS 2024 caste supplement (Part 4b): Individual-level caste \(\times\) industry for SC, ST, OBC, General — enables caste-specific RCA and ECI computation.
• IPL ball-by-ball data (Part 5): Kaggle, 2008–2024, ODbL licence. 724 players, 15 franchises, 17 seasons.
• Proximity Measure: Co-location relatedness across all Indian districts (Hidalgo et al. 2007)
• Informality Definition: Hybrid NCEUS/ILO classification combining employment status, social security coverage, and enterprise size
• ECI Algorithms: Eigenvalue (Hidalgo & Hausmann 2009), Fitness (Tacchella et al. 2012), GENEPY (Sciarra et al. 2020), ECI+ (Albeaik et al. 2017)

References

• Hidalgo, C. A., Klinger, B., Barabasi, A. L., & Hausmann, R. (2007). The Product Space Conditions the Development of Nations. Science, 317(5837), 482–487.
• Hausmann, R., & Hidalgo, C. A. (2011). The Atlas of Economic Complexity. MIT Press.
• Tacchella, A., Cristelli, M., Caldarelli, G., Gabrielli, A., & Pietronero, L. (2012). A New Metrics for Countries’ Fitness and Products’ Complexity. Scientific Reports, 2, 723.
• Sciarra, C., Chiarotti, G., Laio, F., & Ridolfi, L. (2020). Reconciling Contrasting Views on Economic Complexity. Nature Communications, 11, 3352.
• Albeaik, S., Kaltenberg, M., Alsaleh, M., & Hidalgo, C. A. (2017). Measuring the Knowledge Intensity of Economies with an Improved Measure of Economic Complexity. arXiv:1707.05826.
• Asher, S., Lunt, T., Matsuura, R., & Novosad, P. (2021). The Socioeconomic High-resolution Rural-Urban Geographic Dataset on India (SHRUG). Working paper.
• Neffke, F., Henning, M., & Boschma, R. (2011). How Do Regions Diversify over Time? Industry Relatedness and the Development of New Growth Paths in Regions. Economic Geography, 87(3), 237–265.
• Balland, P. A., Boschma, R., Crespo, J., & Rigby, D. L. (2019). Smart Specialization Policy in the European Union: Relatedness, Knowledge Complexity and Regional Diversification. Regional Studies, 53(9), 1252–1268.
• Darity, W. A. Jr. (2015). Stratification Economics: Context versus Culture and the Reparations Controversy. In Position and Responsibility.
• Thorat, S. & Newman, K. S. (2010). Blocked by Caste: Economic Discrimination in Modern India. Oxford University Press.
• NCEUS (2004). Report on Conditions of Work and Promotion of Livelihoods in the Unorganised Sector. Government of India.
• World Bank (2021). The Long Shadow of Informality: Challenges and Policies.

Code & Replication

The urban economic complexity analysis (Parts 1–3) is implemented in Python using PLFS 2024 unit-level data. The district-level panel analysis (Part 4) uses SHRUG Economic Census data with reusable complexity functions (compute_rca, compute_proximity, compute_eci, compute_genepy, compute_eci_plus). The IPL Player Space (Part 5) uses ball-by-ball data from Kaggle (IPL 2008–2024, ODbL licence) covering 724 players across 15 franchises and 17 seasons. All code computes RCA, co-location proximity, density, and SWOT classification from raw microdata. Interactive HTML versions are available upon request.