What Does Venture Capital Actually Finance? Scale and Structure around the World

Santiago Reyes

Kianna Freeman

Pablo Hernando-Kaminsky

Jeremias Huber

Paolo Mauro

Policy Research Working Paper 11438

Abstract

Venture capital is widely viewed as financing knowledge-intensive, R&D-driven startups, but owing to data constraints, this view has been based largely on high-income economies. By taking a global perspective, this paper documents that what venture capital finances difers systematically across economies, varying with institutional and business conditions rather than following a single model. Using a new cross-country dataset that harmonizes firm-level venture capital records with equity issuance data for more than 150 economies, the data shows that VC markets difer across countries not only in scale but in kind. Outside high-income countries, rather than being concentrated in knowledge intangibles, venture capital tilts–at both the

sector and firm levels–toward organizational intangibles such as distribution, logistics, and payments. Highlighting the unique features of venture capital, this pattern is absent from public equity markets, where sectoral composition is far more similar across income levels. An accounting decomposition separates venture capital depth into the rate at which firms enter the market and the funding each entrant attracts, and entry accounts for the largest cross-country gaps. Moreover, the business environment is associated with VC primarily through entry, linked both to the size of the market and to the composition of what venture capital finances, the latter through entry into knowledge-intensive sectors in particular.

What Does Venture Capital Actually Finance? Scale and Structure around the World*

Santiago Reyes^a, ^b Kianna Freeman^a Pablo Hernando-Kaminsky^a Jeremias Huber^a Paolo Mauro^a, ^b

Authorized for distribution Paolo Mauro, Former Deputy Chief Economist and Director of Private Markets Group, Development Economics, World Bank Group

Keywords: Venture Capital, Developing Economies, Sectoral Composition, Startup Financing, Equity Markets, Intangible Capital, Organizational Innovation, Financial Development.

JEL Classi ications: G24, O16, O33, L26, O57

1 Introduction

Venture capital (VC) plays a central role in financing young, innovative firms, and a large body of evidence links VC-backed companies to innovation, firm creation, and economic growth (Kortum and Lerner, 2000; Samila and Sorenson, 2011; Gornall and Strebulaev, 2021). VC has come to be characterized by a particular kind of innovation: knowledge-intensive and R&D-driven, with VC markets elsewhere seen as smaller versions of the same model.¹ Yet the evidence on the type of innovation VC finances comes mostly from the United States (US) and a handful of other high-income economies (Gompers and Lerner, 2001; Hellmann and Puri, 2002; Hochberg et al., 2007; Lerner and Nanda, 2020). Understanding whether this view holds more broadly has become increasingly important, given the rising share of global venture capital flowing to emerging market economies. Moreover, as we document, VC has become the dominant equity financing channel in many developing economies, accounting for roughly two-thirds of total equity financing in the average low- and lower-middle-income country. In this paper, we take a global perspective on what VC finances. Using a newly constructed cross-country dataset that harmonizes venture capital investments recorded in PitchBook and Crunchbase across more than 150 economies, we show that VC difers across countries not only in scale but in kind. The knowledge-intensive investment treated as characteristic of VC is one case rather than the rule: the type of intangibles it finances varies with institutional and business conditions.

The sectoral composition of VC difers systematically across economies. In high-income economies VC concentrates in knowledge intangibles tied to R&D and frontier technology. Moving down the ladder of economic and institutional development, however, VC tilts toward organizational intangibles such as distribution, logistics, and payments. The distinction is established in the intangible-capital literature, which separates knowledge accumulated through R&D, from organization capital, built through management practices, brands, distribution systems, and the human capital embedded in them (Peters and Taylor, 2017; Lev and Radhakrishnan, 2005; Corrado et al., 2009; Ewens et al., 2025). Neither is collateralizable, so both depend more on equity rather than debt, and they difer in what they require of the environment around the firm. A third group sits outside this distinction. Sectors whose value derives mainly from tangible assets, such as agriculture, land transport, and construction account for a larger share of VC in lower-income economies, consistent with VC filling gaps that other sources of capital serve in deeper markets. This variation is specific to VC: public equity markets finance a broadly similar sectoral profile across income levels, whereas the composition of VC shifts with institutional and business conditions.

The tilt towards organizational-intangibles in developing markets is not driven by major emerging venture capital markets like India or China, nor by a few large deals in small VC markets. It is also visible in firms’ own descriptions of what they do, even within narrowly defined sectors like IT. Startups in developing economies are more likely to describe their business in terms of distribution, logistics, and marketplaces, and less likely to reference R&D and frontier technology, than startups in high-income economies.² The pattern therefore does not rest only on sector-level intangible classifications.

This compositional diference is connected to how VC markets develop across countries. An accounting decomposition of VC intensity separates cross-country variation into two margins: entry, measured as the number of startups receiving seed funding per capita, and funding depth per entrant, which captures diferences in ticket sizes and follow-on funding. Entry accounts for the largest cross-country gaps, difering by a factor of 11 to 40 across income groups. Prior work ties institutional quality, tax structure, and equity market depth to aggregate VC activity (Gompers and Lerner, 1998; Jeng and Wells, 2000; Cumming et al., 2010; Groh and Wallmeroth, 2016). We show that these conditions act on VC unevenly: their association runs largely through entry rather than funding depth, and it is strongest for entry into knowledge-intensive sectors. The business environment is thus linked not only to how much VC a country has, but to what that VC finances. This is consistent with knowledge-intensive ventures requiring longer time horizons, higher tolerance for failure, and stronger institutional safeguards, leaving them more exposed to institutional weaknesses than organizational ventures that generate returns more quickly. The conditions associated with VC market depth are therefore also associated with its sectoral composition, one channel consistent with the organizational orientation we document in developing economies.

These questions have received limited attention in part because no single data source provides comprehensive coverage of VC in developing economies. We make two contributions on this front. First, we construct a harmonized dataset that combines the two largest VC databases (PitchBook and Crunch base) into a unified firm-level dataset through a multi-stage matching procedure that links firms across sources without a common identifier. Throughout the paper, we define VC financing broadly to include pre-seed, seed, angel, and venture capital rounds. The combined dataset captures 15%–42% more VC activity than either PitchBook or Crunchbase alone, depending on the measure used, with the largest gains occurring in developing economies. This expanded coverage is not merely additive: it provides a more complete picture of the scale and sectoral composition of VC activity in regions where PitchBook and Crunchbase individually miss a substantial share of deals. Second, we standardize sector classifications across datasets using ISIC codes³ and match the harmonized VC data with country-sector-level equity issuance data from the London Stock Exchange Group (LSEG).⁴ This allows us to situate VC within each country’s equity market structure, to compare what VC finances against public equity markets, and to measure how far VC in each economy departs from the high-income model. The resulting dataset is the first to combine firm-level VC data with sector-level equity issuance across this range of economies.

These findings speak to four areas of research. First, the paper extends our understanding of what VC finances globally. Reflecting the rise of VC and private equity across the world, there is growing interest in analyzing these markets in a wider range of countries. Focusing on China, Lerner et al. (2026) suggest that the difusion of VC into emerging markets depends on the contextual appropriateness of business models. Colonnelli et al. (2025) point to the rapid expansion of private capital markets in emerging and developing economies and the wide variation of sectoral and deal-size patterns in such markets, as a prelude to analyzing investor attitudes measured through survey experiments with private capital investors. Colonnelli et al. (2026) study the demand side in Africa, documenting startups’ financing preferences and the foreign dominance of local equity supply. We focus on venture capital in the strict sense, the equity financing of young firms rather than private equity or buyouts. Our paper does not just describe which sectors VC finances across countries; it asks what kind of intangible capital it finances in diferent institutional settings. The harmonized dataset reveals that the knowledge-intensive characterization of VC is a feature of high-income markets. VC in developing economies finances a diferent type of intangible investment, one oriented toward management capabilities, distribution systems, and market access rather than R&D and patenting. This shift in intangible composition is specific to VC and it has impli cations for whether the innovation spillovers documented in high-income settings (Kortum and Lerner, 2000; Samila and Sorenson, 2011; Popov and Roosenboom, 2012) can be expected to operate where VC finances diferent activities.

Second, the composition finding connects to a broader insight in the entrepreneurship literature: that institutions shape not just how much entrepreneurship a country produces but what kind (Baumol, 1996; Webb et al., 2020). Our results provide large-scale evidence for the financing side of this argument. Where institutions are weaker, VC finances organizational innovation rather than knowledge-based innovation, consistent with entrepreneurs pursuing opportunities that require less institutional support (Estrin et al., 2013; Ahlstrom and Bruton, 2006). The tilt toward distribution, logistics, and payments is also consistent with VC-backed firms building intermediary infrastructure where institutional voids leave these functions under-served (Khanna and Palepu, 1997).

Third, the paper extends the literature on cross-country VC determinants, which began with U.S. evidence on fundraising (Gompers and Lerner, 1998) and was subsequently extended to multi-country settings (Jeng and Wells, 2000; Cumming et al., 2010; Aizenman and Kendall, 2012; Groh and Wallmeroth, 2016). Groh and Wallmeroth (2016) find that standard determinants show systematically diferent magnitudes in emerging markets. We introduce an accounting identity that decomposes VC intensity into entry and funding depth per entrant. Because the identity is additive in logs, it permits an exact attribution of cross-country variance to each margin, revealing that the dominant margin difers across income levels. It also allows us to assess how the association between business-environment features and VC intensity is distributed across margins. The diferential sensitivity of knowledge-intensive entry to institutiona conditions is consistent with the sectoral composition gap we document.

Fourth, the paper contributes to the study of how financial structure shapes economic development (Levine, 1997; Rajan and Zingales, 1998; Beck et al., 2000). Standard cross-country measures of financial development track banking depth and stock market capitalization but do not capture venture capital. We show that this omission is consequential: in developing economies, VC accounts for roughly two-thirds of total equity financing in the average low- and lower-middle-income country, making it the primary channel through which innovative firms access equity capital. Moreover, VC’s sectoral composition difers from that of public equity financing, and it shifts with income level in a way that the composition of equity financing does not. Aggregate measures of financial depth cannot capture this, since they treat financing as homogeneous. This connects to the growing body of work on capital markets in low- and middle-income countries (Meh and Schmukler, 2025), which documents that these markets are not simply underdeveloped versions of high-income financial systems but serve diferent functions within their economies.

The paper is structured as follows. Section 2 describes the data sources and harmonization process. Section 3 presents the evidence, organized around two dimensions: the sectoral composition of VC across income levels, and the development of VC markets across countries, including the conditions associated with cross-country variation. Section 4 discusses implications for policy and future research.

2 Data

The central data challenge in measuring VC activity outside the traditional hubs is that no single source provides comprehensive coverage. We address this by harmonizing PitchBook and Crunchbase into a unified firm-level dataset and matching it with LSEG equity issuance data at the country-sector level.⁵ This section describes the underlying sources, the harmonization process, and the gains from combining the two VC datasets.

2.1 Data Sources

Venture Capital. We draw on the two largest global databases on venture capital, PitchBook and Crunchbase, to construct the harmonized dataset. Both capture a broad range of deal types across the firm life cycle, from early-stage financing to IPOs, with coverage spanning all regions (see Appendix B for additional details on each dataset). Neither source, however, is comprehensive on its own, particularly in developing economies where deal-level data are harder to collect and verify. This limitation has been documented elsewhere. Kaplan and Lerner (2017) note that no single source captures every venture transaction, and Colonnelli et al. (2026) show that supplementing PitchBook with specialized local providers in Africa recovers 39 percent more deals and 25 percent more companies. We exploit the complemen tarity between the two largest global databases to construct a coverage-gain measure consistently across all countries in our sample. As we document below, these gains are largest in middle-income countries, where the provider with the most complete coverage varies across countries.

Table B.1 in the appendix summarizes the raw coverage of each source. PitchBook is the larger database overall, with 561,918 firms, but much of its coverage extends to non-VC financing like buyouts or private equity. Crunchbase is more specialized: 91.9% of its firms have received venture capital financing, compared to 47.7% in PitchBook. We define VC financing to include pre-seed, seed, angel, and venture capital rounds. Consequently, we restrict the sample to firms that received at least one such round and, to reduce measurement error, further limit it to firms that received their first deal before age 15.⁶

We also classify firms into International Standard Industrial Classification (ISIC) sectors to make the venture capital data comparable with other firm- and market-level datasets. Sector information in PitchBook and Crunchbase is reported using provider-specific industry categories, which are not directly comparable across sources or with external datasets. We therefore map firms into ISIC codes using the available sector and industry fields. This harmonization is important for linking VC activity to other datasets, such as LSEG equity issuance data in this paper, and for comparing how sectoral patterns difer across venture capital, public equity, and other sources of firm financing.

The empirical analysis focuses on deals recorded between 2012 and 2023, a period during which venture capital activity is observed more consistently across all regions. For the analysis, we group deals into three-year windows: 2012–2014, 2015–2017, 2018–2020, and 2021–2023 to reduce year-to-year volatility. The dataset covers investments by both private and public investors since our goal is to characterize the overall structure of VC markets across countries rather than to distinguish between investor types.

Equity Issuance. We match the harmonized VC data with country-sector-level equity issuance data constructed from the Securities Data Company (SDC) Platinum database provided by the London Stock Exchange Group (LSEG) (see Appendix B for additional details). SDC Platinum provides detailed transaction-level information on equity issuances by publicly listed and privately held firms across countries, enabling us to document and characterize equity market activity worldwide. Throughout, we assign firms to countries by the location of their headquarters rather than the venue of issuance or listing, the same basis on which we assign venture capital. A firm headquartered in Colombia that lists abroad is therefore recorded as Colombian equity issuance, just as a Colombian-headquartered firm funded by a foreign venture investor is recorded as Colombian VC. This common basis is what allows the two financing channels to be compared at the firm level and lets us benchmark VC activity against public equity markets at the country-sector level.

Private Equity Growth. Private equity growth is measured using PitchBook data by retaining deals with deal type classified as private equity. Deals are assigned to three periods — 2015–2017, 2018–2020, and 2021–2023 — and observations with zero or missing deal size are excluded. Deal volume is then summed by country and period, producing a country-period measure of private equity growth activity.

Other indicators. Countries are assigned to income groups and regions using the World Bank’s 2025 classifications (see Appendix C). Country-level measures of the business environment are drawn from several sources: minority investor protection from the World Bank’s Doing Business dataset; political stability from the World Bank’s Worldwide Governance Indicators; and tax-related measures, including tax loss carryforward allowance, the efective capital gains tax rate, and the corporate income tax rate, from PricewaterhouseCoopers (PwC).

2.2 Harmonization Process

The central challenge in combining PitchBook and Crunchbase is distinguishing firms that appear in both datasets from those unique to one source. Two types of error can arise in this process. First, a false negative would occur if a firm exists in both datasets but is not detected as a match; this leads to double counting and inflates deal volumes. Second, a false positive would occur if two records are matched but refer to diferent firms. This can result in the loss of valid observations and understate activity. No universal firm identifier links the two sources. To minimize those problems, we use a multi-stage matching procedure that combines website-based matching, firm-name matching, and a large language model (LLM)-based verification of candidate matches using firm descriptions and sectoral classifications.

The procedure consists of five stages.⁷ In the first stage, we standardize website URLs across both datasets and match firms within each country using the cleaned URLs. Perfect matches, defined as pairs with a match score equal to 1, bypass the intermediate verification stages and proceed to the fifth stage for validation, while candidate pairs with match scores between 0.80 and less than 1 proceed to the second stage. In the second stage, we use an LLM to verify whether each non-perfect website candidate pair refers to the same entity, drawing primarily on firm descriptions. Confirmed matches are retained and proceed to the fifth stage, while unconfirmed pairs continue to the third stage. Stages three and four repeat this process using standardized firm names rather than URLs: Stage 3 generates perfect and non-perfect firm-name candidate matches, and Stage 4 uses LLM-based description verification for the non-perfect firm-name matches. In the fifth stage, we select the 60 largest matched firms in each country-dataset, ranked by the total amount of venture capital received by the firm, and conduct an additional round of LLM-based verification. This final stage covers both perfect matches and non-perfect matches confirmed in earlier stages; matches confirmed through this review are retained, while rejected pairs are returned to the unmatched pool.

Validation. The validation exercise highlights the tradeof involved in selecting match thresholds. Table A.4 in the appendix shows that fuzzy matching below a score of 1.0 recovers genuine matches that would otherwise be missed because of minor diferences in URLs or firm names. However, confirmation rates fall sharply below 0.95, and lowering the threshold below 0.80 yields few additional matches at high computational cost. The fifth stage review also shows that perfect string matches are not necessarily error-free. Among firms with identical URLs included in the top validation sample, 2.2% were found to refer to diferent entities after their firm descriptions were compared.

The final dataset consists of 269,915 startups that received at least one deal between 2012 and 2023.⁸ Of these, 20% (54,836) appear only in Crunchbase, 32% (86,641) appear only in PitchBook, and 48% (129,900) are matched across both sources. For firms matched across the two databases, we retain the PitchBook record as the primary data source, including its deal history, sector classification, location, and firm description.

2.3 Gains from Harmonization

Combining PitchBook and Crunchbase into a single dataset generates large gains in coverage, and these gains are concentrated in the markets where existing data are weakest. Panel (a) of Figure 1 shows the increase in observed VC volume relative to using PitchBook alone. The gains are largest in upper-middle income countries, where harmonization adds 22.2% more volume for the median country and 17.4% in the aggregate. Low- and lower-middle-income countries also gain substantially: 16.8% for the median country and 15.4% in the aggregate. In high-income countries, the US, and China, the gains are smaller, consistent with the higher visibility of deals in these markets and more comprehensive coverage in both Pitchbook and Crunchbase.

The regional breakdown in Panel (b) makes the pattern sharper. Sub-Saharan Africa gains the most in volume, adding 29.7%, while Latin America and the Caribbean gains the most in deal counts, adding 42.3%. These are the regions where single-source data most understate VC activity, so cross-country comparisons drawn from either source alone would most distort the position of those regions. The gains are not merely additive: the unique firms contributed by each source show that the datasets capture distinct segments of VC activity in developing economies. PitchBook adds more unique firms from highincome countries, while Crunchbase adds more from upper-middle-income, lower-middle-income, and low-income countries. The sectoral composition of these unique contributions also difers, meaning that harmonization improves not only the scale of coverage but also its representativeness across sectors and economic contexts.

Figure 1: Gains in Venture Capital Data Coverage
(a) Gains in Total USD Volume by Income Group

(b) Gains in Total USD Volume and Number of Deals by Region

Source: Harmonized PitchBook and Crunchbase dataset.
Note: This figure shows the gains from harmonization in venture capital activity relative to PitchBook, by income group and region. Values are expressed relative to the PitchBook baseline, where 100 equals total volume in PitchBook, so values above 100 indicate gains from harmonization. Panel (a) reports gains in total VC volume by income group, along with separate values for China and the United States. Panel (b) reports gains in total VC volume and the number of deals by region. Income group and regional classifications follow the World Bank’s 2025 classifications.
Regional abbreviations in panel (b) are as follows: SSA = Sub-Saharan Africa, LAC = Latin America and the Caribbean, EAP = East Asia and Pacific, MENA = Middle East and North Africa, SAR = South Asia, and ECA = Europe and Central Asia.

Table B.5 in Appendix B presents descriptive statistics for the harmonized dataset. The final sample contains 527,031 deals across 269,915 firms, of which 436,273 deals (215,729 firms) are in high-income countries and 90,758 deals (54,186 firms) are in non-high-income economies. The distribution of deal sizes difers across income groups: developing countries have a higher share of deals below \$100,000 (31.5% versus 19.4%) and fewer deals above \$3 million (17.2% versus 25.5%). China is a particular case, where even the smallest deals clear the \$100,000 threshold. The probability of receiving follow-on investment is 36.7% in non-high-income countries compared to 45.9% in high-income economies.⁹ The regional distribution of firms reflects the well-known concentration of VC activity: North America accounts for 53.8% of firms in the high-income sample, while East Asia and Pacific dominates the non-high-income sample at 52.8%.

The sectoral composition of the dataset foreshadows the patterns we document in Section 3.1. IT is the largest sector in both groups but accounts for a smaller share of firms in non-high-income countries (41.0%) than in high-income economies (47.8%). Non-high-income countries have a larger share of firms in wholesale and retail (8.1% versus 4.0%), transportation and logistics (2.9% versus 1.2%), agriculture and natural resources (2.8% versus 2.3%), and financial services (4.5% versus 3.2%). This broader sectoral spread is consistent with VC in developing economies responding to local market opportunities rather than replicating the IT-heavy composition typical of high-income VC markets.

3 How Venture Capital Markets Take Shape

Most of what is known about venture capital comes from high-income economies. The one well-established fact about VC in developing economies is that it is small relative to GDP, on average roughly five times smaller than in high-income countries¹⁰. Beyond this, systematic evidence on the role VC plays within the broader financial system, and what it finances is limited. Yet the GDP benchmark, while informative about scale, says little about VC’s role relative to the financing options actually available to firms. Measured as a share of total equity financing, combining our harmonized dataset with equity issuance and private equity growth data from LSEG, the picture reverses. VC accounts for 66.3% of equity financing in the average low- and lower-middle-income country in 2021–2023, compared to 37.7% in the average high-income economy (Figure 2a). This share has risen sharply over the past decade: in 2015–2017, the average stood at 37.9% in low- and lower-middle-income countries and 24.3% in high-income economies. Part of this reflects the thinness of the alternatives: in some of these economies, public equity issuance and private equity growth are negligible or absent, so VC is efectively the only equity financing channel available to firms. VC has grown faster than public equity issuance and private growth equity combined across all income groups (Figure 2b), but the shift is sharpest in developing economies, where the average VC share of equity activity nearly doubled over six years.

Black and Gilson (1998) argue that active public equity markets are a precondition for functioning VC markets. The cross-country pattern suggests that the relationship may also run in the other direction: where public equity and private equity remain shallow, VC serves as the dominant alternative. In lowincome countries, policies afecting VC markets would therefore not shape a marginal segment of the financial system. They would shape the primary or only channel through which firms access equity capital. This makes the composition of VC, and the factors that determine its depth, first-order questions for understanding how developing economies finance innovation. We examine both questions in the remainder of this section.

Figure 2: The Role of VC in Equity Markets Beyond High-Income Countries
(a) Average Share of VC in Equity Financing

(b) Cumulative Growth in VC Volume

Sources: Harmonized PitchBook and Crunchbase dataset and equity data from calculations using issuance data from the Securities Data Company Platinum database from LSEG.

Notes: This figure shows two measures of VC activity by income group. Panel (a) reports the average share by income groups of VC in equity markets, defined as the total value of VC deals divided by the sum of VC, public equity issuance, and private equity growth investment, for three sub-periods between 2015 and 2023. The average includes countries where either VC or equity issuance is zero, as long as it has some type of equity financing. Panel (b) reports cumulative growth in total VC volume by income group relative to the 2012–2014 baseline period, with the series normalized to zero in 2012–2014. The figure uses the harmonized PitchBook–Crunchbase dataset and focuses on VC-backed young firms. Country groupings follow the World Bank’s 2025 income classifications.

3.1 The Sectoral Composition of Venture Capital

Treating VC volume as a single aggregate, implicitly assuming that a dollar of VC finances similar activities regardless of where it is deployed. If this were true, VC markets in developing economies would simply be smaller versions of high income markets, with the same sectoral allocation scaled down. The granularity of the harmonized dataset allows us to test this assumption directly. We classify each deal into 17 sectors based on firm’s ISIC classification and compute, for every country, the cosine similarity between its sectoral VC distribution and that of the US, the largest VC market.¹¹ Figure 3 maps the results. The sectoral composition of VC varies widely across countries, and divergence from the US pattern rises as income falls. This divergence is specific to VC. Equity issuance markets look far more similar to the United States across the income distribution: the median country has a cosine similarity of 0.82 for equity issuance, against 0.58 for VC (Figure E.3, Appendix E). The contrast matters because the development implications of VC depend on what it finances. A VC market concentrated in IT and software raises diferent questions for industrial policy, human capital, and growth than one concentrated in agriculture, logistics, or financial services. If developing economies are not simply following the US template, the natural question is what VC actually finances in these contexts, and what that reveals about its role at diferent stages of development.

Figure 3: Cosine Similarity Index relative to US Sectoral Distribution¹²

Sources: Harmonized PitchBook and Crunchbase dataset.

Notes: This figure shows the cosine similarity between each country’s sectoral distribution of venture capital activity and that of the United States. The index ranges from 0 to 1, with higher values indicating that a country’s VC sectoral mix is more similar to the U.S. benchmark and lower values indicating a more distinct sectoral composition. Darker blue shades denote greater similarity, lighter blue shades denote lower similarity, and gray indicates countries with no available data. Sectoral groupings are based on ISIC classifications.

The sectoral distributions behind these patterns make the divergence concrete. Figure 4 shows the average share of VC volumes by sector across high-income countries and developing markets. In high-income countries, VC is heavily concentrated in IT, which accounts for 40.8% of total volume, with sectors like R&D services and chemicals and pharmaceuticals manufacturing also representing a larger share relative to developing economies. In low- and middle-income countries, IT remains the largest sector at 29.7%, but venture capital is spread across a broader set of sectors. Wholesale and retail captures 14.6% of VC volume, compared to 5.8% in high-income countries. Transportation and logistics accounts for 8.0%, nearly three times the high-income share of 2.9%. Financial services, other manufacturing, and education also account for larger shares of VC in developing economies than in high-income countries. Rather than replicating the high-income sectoral mix on a smaller scale, investors in these markets are allocating capital to a diferent set of opportunities.

We have shown that the sectoral allocation of VC difers across income levels. A related question is whether the nature of the activity being financed also difers. Venture capital plays a critical role in financing intangible-intensive firms whose value is embedded in knowledge, software, intellectual property, and organizational capabilities. Because such firms often lack collateralizable assets, operate under substantial uncertainty, and are subject to pronounced information asymmetries, they face significant barriers to obtaining financing through conventional debt markets (Hall and Lerner, 2010; Gompers and Lerner, 2001). Moreover, these characteristics increase the returns to investor monitoring and value-added support, further reinforcing the comparative advantage of venture capital investors.

Figure 4: Distribution of VC across sectors and Income groups

Sources: Harmonized PitchBook and Crunchbase dataset and GDP data from World Bank’s World Development Indicators. Note: This figure compares the average sectoral distribution of venture capital volume within countries across low- and middle-income countries and high-income countries. For each income group, the figure reports the average share of VC volume accounted for by each sector over 2012–2023, weighted by country GDP. Sectoral groupings are based on ISIC classifications, and income classifications are based on the World Bank’s 2025 income classifications. Values are expressed as percentages of total VC volume within countries.

We test whether this focus on intangibles holds across income levels using the sector-level intangible intensity classification from Demmou et al. (2020), which builds on the firm-level framework in Peters and Taylor (2017). Peters and Taylor decompose intangible capital into two components: knowledge capital, measured as accumulated R&D spending, and organization capital, captured by a fraction of selling, general, and administrative expenses that reflects management practices, brands, distribution systems, and human capital investments. A recent literature treats sales-and-marketing spending as investment in customer capital — the distribution networks, market access, and demand-side relationships whose value is largest in industries characterized by search frictions and network efects (He et al., 2024). Demmou et al. (2020) translates these firm-level measures into sector-level averages, allowing us to classify each sector according to their intensity in three types of intangibles: total intangibles, knowledge-based intangibles, and organization-based intangibles (see Appendix D for complete sectoral classification). These proxies of knowledge and organizational capital depend on assumed capitalization parameters, and may understate the true value of intangible assets (Ewens et al., 2025). Although the literature on intangibles, and therefore its sectoral classification, has been largely developed using firm-level data from high-income countries (Corrado et al., 2009; Peters and Taylor, 2017), there are strong theoretical reasons to expect the classification to remain valid in developing economies. Intangible intensity is primarily a property of a sector’s production technology rather than its geographic location: IT, for instance, requires coinvestment in organizational processes, worker training, and decentralized decision-making regardless of where the firm operates (Brynjolfsson and Hitt, 2003). Furthermore, intangible assets are non-rival — once developed, they can be deployed across borders without being depleted (Corrado et al., 2022), meaning that firms in developing countries that deploy existing technologies inherit the intangible intensity of those technologies.¹³

At first glance, the standard characterization holds. Across all countries, VC allocates a larger share of volume to intangible-intensive sectors (Table 1, column 1). The association is weaker in developing economies, where the share going to intangible-intensive sectors is lower than in high-income countries (column 2), but it remains positive. This aggregate result, however, masks an important distinction. If the definition of intangibles is based solely on knowledge intangibles, the patterns diverge. The concentration of VC in knowledge-intensive sectors, those defined by R&D and patenting spending, is driven almost entirely by high-income countries (columns 3–4). This is the setting where Kortum and Lerner (2000) document a strong link between VC activity and patenting rates. In developing countries, the net association between VC volume shares and knowledge intangible intensity is not significantly diferent from zero (p = 0.34 for low- and lower-middle-income, p = 0.81 for upper-middle-income countries).

VC in developing economies does tilt toward intangibles if the definition expands to include organizational intangibles: those defined by management capabilities, distribution networks, brand building, and human capital (columns 5–6). The net association is weaker than in high-income countries, but it is positive and significant across all income groups. This distinction is consistent with what VC-backed firms in developing economies actually do. Companies like Rappi in Colombia or M-Pesa-era mobile banking platforms in East Africa did not develop new core technologies; they applied existing technologies to distribution, payments, and logistics problems specific to underserved markets. Their intangible value lies not in patents or R&D but in the organizational knowledge required to operate in contexts where infrastructure is limited and consumer needs difer from those in high-income markets. This interpretation is consistent with evidence from Africa that founders value investors with local knowledge (Colonnelli et al., 2026). The standard narrative that VC finances innovation through intangible capital is not wrong, but it is incomplete. In developing economies, the relevant intangibles are organizational rather than knowledge-based, and the innovation lies in deployment rather than discovery. Evidence on firm dynamics in emerging markets shows that high-growth firms are not concentrated in knowledge-intensive sectors; they span services, retail, and manufacturing, and account for a disproportionate share of job creation regardless of sector (Grover et al., 2019).

A final fact from Table 1 concerns the share of VC going to the base group: sectors classified as nonintangible in the total measure (column 1), and the same sectors in high-income countries (column 2). The constant indicates that these sectors receive a non-trivial share of VC volume (11%) in high-income countries, and the observed shares is higher in developing economies. We refer to these as traditiona sectors, in the sense that their value derives primarily from tangible assets (see Appendix D for complete sectoral classification). They include activities such as agriculture and crop production, land transport, basic utilities, and construction. The label describes their position in the intangibles classification, not their growth potential, and their presence raises a question. These are firms whose assets are precisely the kind banks lend against, yet they raise venture equity rather than debt. Why VC finances activity that the traditional theory assigns to bank lending is a pattern we discuss in subsection 3.2.

Table 1: Intangible Intensity and VC Volume Shares (2012–2023)

	(1)	(2)	(3)	(4)	(5)	(6)
	(1)	Share of VC Volume by intangible definition		Knowledge		Organizational
High intangible intensity (total)	0.561*** (0.022)	Total 0.617* -0.229*
× Upper-middle income							(0.016) -0.144*** (0.046)
× Low & lower-middle income						0.158***
High intangible intensity (knowledge)		(0.045)	(0.030)	0.239* (0.022) -0.224*
× Upper-middle income × Low & lower-middle income				(0.067) -0.276***		0.579***
High intangible intensity (organizational)				(0.044)	0.532***	(0.013)
× Upper-middle income					(0.019)	-0.128*** (0.034)
× Low & lower-middle income					0.149***	-0.165*** (0.044)
Constant	0.134*** (0.012)	0.111*** (0.007)	0.336*** (0.018)	0.300*** (0.013)	(0.012)	0.130*** (0.005)
Observations R-squared	483 0.868	483 0.887	483 0.591	483 0.675	483 0.875	483 0.890
p-value: Low & LMI = 0		0.000		0.338		0.000
		0.000		0.809		0.000
p-value: Upper-middle = 0

Notes: This table reports regressions of the share of VC volume (2012–2023) allocated across sectors on measures of sectoral intangible intensity. The dependent variable is the share of VC volume in each sector-country cell. In each column, three types of sectors are formed: those where the definition holds, those where the definition does not hold, and those without data for classification. The table reports data for the first two. Columns (1) and (2) use the overall intangible intensity measure, columns (3) and (4) use the knowledge intangible intensity measure, and columns (5) and (6) use the organizational intangible intensity measure. Even-numbered columns interact intangible intensity with indicators for low- and lower-middle-income countries and upper-middle-income countries, so that high-income countries are the omitted category. Income classifications are based on the World Bank’s 2025 income classification. Regressions are weighted by country GDP. Standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

We repeat the exercise for public equity issuance using LSEG data. The results demonstrate the distinctiveness of VC. Equity issuance does not favor intangible-intensive sectors. If anything, it tilts toward sectors with lower intangible intensity (Table 2). More telling is that this pattern does not vary consistently across income levels. Unlike VC, where the type of intangible activity financed difers between developing and high-income economies, public equity markets finance a similar sectoral profile every where. This is consistent with Lerner et al. (2026), who suggest that the difusion of venture activity across emerging markets was driven not by capital flows but by local investors adapting business models to conditions specific to each context.

Table 2: Intangible Intensity and Equity Volume Shares (2012–2023)

	(1) Total	(2)	(3)	(4) Knowledge Intangible Int.	(5) Organizational Intangible Int.	(6)
High intangible intensity (total)	-0.230***	-0.197***
× Low & lower-middle income	(0.039)	(0.032) -0.357***
× Upper-middle income		(0.040) -0.033
High intangible intensity (knowledge)		(0.116)	-0.673***	-0.700***
× Low & lower-middle income			(0.046)	(0.049) -0.048
× Upper-middle income				(0.056) 0.111
High intangible intensity (organizational)				(0.093)	0.017	-0.019
× Low & lower-middle income					(0.040)	(0.040) 0.039
× Upper-middle income						(0.066) 0.123
Constant	0.612***	0.599***	0.834***	0.850***	0.489***	(0.095) 0.509***
Observations	(0.027)	(0.023)	(0.032)	(0.035)	(0.027)	(0.028)
R-squared	244	244	244	244	244	244
p-value: Low & LMI = 0	0.333	0.383	0.861	0.866	0.003	0.030
p-value: Upper-middle = 0		0.000 0.040		0.000 0.000		0.710 0.230

Sources: Equity data from calculations using issuance data from the Securities Data Company Platinum database from LSEG. Notes: This table reports regressions of the share of equity volume allocated across sectors on measures of sectoral intangible intensity. The dependent variable is the within-country share of equity volume. Columns (1)–(2) use the overall intangible intensity measure, columns (3)–(4) use the knowledge intangible intensity measure, and columns (5)–(6) use the organizational intangible intensity measure. Even-numbered columns interact intangible intensity with indicators for low- and lower-middle-income countries and upper-middleincome countries, so that high-income countries are the omitted category. Income classifications are based on World Bank’s 2025 income classification. Regressions are weighted by country GDP. Standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

As mentioned above, a concern with using sector-level classification is that it assigns intangible intensity based on the production technology of each industry, which might difer firm to firm or sectors may difer across income levels. As complementary evidence, we check whether the pattern holds at the firm level. We construct two keyword dictionaries, one for knowledge-oriented activity and one for organizationaloriented activity, by prompting a large language model with the definitions of the two intangible types used above and asking it to generate associated terms, which we then reviewed. The knowledge dictionary covers R&D, patenting, and frontier technology (e.g., proprietary, patent, algorithm, clinical trial, drug discovery); the organizational dictionary covers distribution, logistics, and market access (e.g., logistics, last-mile, marketplace, supply chain, mobile money). We then flag each startup by searching its business description for these terms.¹⁴ This analysis serves as a robustness check on the sector-level classification and is not meant to capture every term associated with each category.

Table 3 regresses the prevalence of each keyword type on income group indicators, with high-income countries as the omitted category. The results confirm the sectoral patterns. In high-income countries, 16.0% of firm descriptions contain at least one knowledge keyword, compared to 10.2% in low- and lower-middleincome economies and 10.1% in upper-middle-income economies — a gap of roughly six percentage points.

The pattern reverses for organizational keywords: 9.3% of high-income firm descriptions contain organi zational terms, compared to 17.7% in low- and lower-middle-income and 16.5% in upper-middle-income economies. Column 3 summarizes the net orientation: high-income VC-backed firms are tilted toward knowledge intangibles relative to organizational ones, while developing-country firms tilt in the opposite direction. The diference in this knowledge-minus-organizational gap between high-income and developing economies is 13–14 percentage points. The distinction between knowledge and organizational intangible orientation is not an artifact of sectoral classification. It is also reflected in firms’ own descriptions of their activities.

Table 3: Knowledge and Organizational Keywords in Business Model

VARIABLES	(1) Knowledge	(2) Organizational	(3) K minus O
Low & lower-middle income	-0.058***	0.084***	-0.143***
Upper-middle income	(0.002) -0.060***	(0.003) 0.073***	(0.004) -0.132***
Constant	(0.003) 0.160***	(0.004) 0.093***	(0.005) 0.068***
	(0.001)	(0.001)	(0.001)
Observations R-squared	240,911	240,911	240,911
	0.003	0.007	0.009

Notes: The table reports firm-level regressions of keyword-based indicators on income-group dummies. The dependent variable in column (1) equals one if the firm’s business description contains at least one knowledge-related keyword, including terms related to R&D, patents, algorithms, artificial intelligence, biotechnology, semiconductors, and frontier technology. The dependent variable in column (2) equals one if the firm’s business description contains at least one organizational-related keyword, including terms related to distribution, logistics, marketplaces, supply chains, delivery networks, customer acquisition, e-commerce, mobile money, and digital payments. Column (3) reports the diference between the knowledge and organizational indicators. Robust standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

3.2 From Sectoral Structure to Venture Capital Depth

The sectoral patterns in subsection 3.1 are not confined to a comparison between high-income and developing economies. They vary continuously with the depth of VC markets themselves. Figure 5a plots, against log VC volume as a share of GDP, the share of VC going to sectors depending on their intangible definition (knowledge, organizational, or neither - traditional). Because the knowledge and organizational measures are not mutually exclusive, a sector can be intensive in both definitions, and these shares need not sum to one. The knowledge and organizational-intensive shares rise steadily with VC depth, while the traditional share declines. The sectoral tilt documented in the previous subsection is one expression of a broader pattern: as VC depth changes across countries, so does what VC finances.

A natural question is how VC fits within the rest of the financial system. The relationship between VC depth and financial development is far from tight. Figure 5b plots VC-to-GDP against the IMF Financial Development Index. The two are positively correlated, but the dispersion is wide among countries scoring below 0.6, where VC-to-GDP ratios span several orders of magnitude at similar levels of financial develop ment. Financial development accounts for roughly one-third of the cross-country variation in VC depth, leaving the remaining two-thirds to other factors. One reading consistent with this dispersion is that VC substitutes for banks where formal credit is scarce: in some low-development settings VC finances activity that deeper banking systems would absorb, which would also explain why traditional, asset-based sectors take a larger share of VC there. These patterns raise the question of what drives cross-country diference in VC development.

These patterns motivate a closer look at the components of VC development, as they shed light on the underlying conditions associated with what VC finances in diferent markets. The rest of this section examines those components. We document the scale of cross-country variation in VC development, decompose it into two margins (the entry into VC markets and the funding each entrant attracts), and ask how much of the cross-country variation, and of its association with business-environment conditions, runs through each.

Figure 6 shows the wide variation in VC development across countries. It maps each country’s VC-to-GDP ratio relative to what its income level would predict ¹⁵. Among the strongest overperformers–—that is, countries whose VC-to-GDP ratios exceed the levels predicted by income— are Israel and Estonia, both high-income economies with well-documented policy interventions that catalyzed their VC markets.¹⁶ But such outstanding performance is not confined to wealthy countries: India, Kenya, and Nigeria all attract VC volumes well above what their income levels alone would suggest. At the other end, countries like Angola, Algeria, and Paraguay fall among the furthest below expectations, with VC activity several standard deviations below what income would predict.

Figure 5: From Sectoral Composition to VC Development
(a) Sector Shares and VC Depth

(b) Correlation VC / GDP and Financial Development

Sources: Harmonized PitchBook and Crunchbase dataset; IMF Financial Development Database; GDP data from World Bank World Development Indicators.

Notes: Panel a plots the share of VC volume allocated to knowledge-intensive, organizational-intensive, and traditional sectors against the log of VC volume as a share of GDP. Each curve is a local polynomial cross-country relationship. Sectoral classifications are based on Demmou et al. (2020): sectors are classified as knowledge-intensive or organizational-intensive if their intangible intensity along each dimension exceeds the cross-sector mean; traditional sectors are those intensive in neither. Panel B plots the relationship between venture capital activity and financial development across countries for 2021–2023. The vertical axis reports the log of VC deal volume as a share of GDP. The horizontal axis reports the IMF Financial Development Index. Each point represents a country, and the fitted line shows the linear relationship between log VC/GDP and the Financial Development Index.

Figure 6: Development of VC relative to expectation based on income level¹⁷

Sources: Harmonized PitchBook and Crunchbase dataset; World Bank World Development Indicators; World Bank geospatial data. Notes: The map shows country-level over- or under-performance in venture capital activity relative to the level predicted based on income during 2021–2023. For each country, VC activity is measured as total VC deal volume relative to GDP. The log of this ratio is winsorized at the 1st and 99th percentiles, rescaled to a 0–100 index, and regressed on log GDP per capita. The mapped variable is the standardized residual from this regression, so positive values indicate VC activity above the predicted level, and negative values indicate VC activity below the predicted level. Countries shown in gray have no data.

To unpack the cross-country diferences in VC development, we use an accounting identity that decomposes VC intensity into economically meaningful components.¹⁸ Let V denote total VC investment in a country, N the total number of deals (from pre-seed to late-stage VC), and t = V/N the average ticket size, so that V = N t. We split deals into entry rounds, pre-seed and seed (N₁), and later-stage deals (N_f). Entry rounds mark a firm’s access to VC, typically to test and validate a business idea before it has the traction that later-stage capital requires. They are small, as low as \$2,500, with 10% below \$20,000. Later rounds, by contrast, can be understood as scale-up financing, provided to firms that have already entered and are growing (see Durufl´e et al. (2017)). As a count, N₁ tracks the extensive margin, how many firms cross into VC at all, while the capital each raises loads onto the intensive margin. Factoring out N₁:

(1)

where t₁ and t_f are the average ticket sizes of seed/pre-seed and later-stage deals, respectively.

Dividing by population (P)¹⁹ yields two economically meaningful margins: entry per capita and funding

depth per entrant

(2)

Taking logs turns the identity additive. Additivity is what makes the decomposition useful beyond accounting. Any linear operation applied to log VC depth passes through to the two margins and splits exactly between them, with the parts summing to the whole. A country’s deviation from any benchmark is the sum of an entry deviation and a depth deviation. The same holds for the association between VC depth and any country characteristic: the coeficient splits exactly into an entry and a funding depth component. These properties let us ask which margin accounts for cross-country diferences in VC development, and whether that margin is also the one that shapes what VC finances.

(3)

We apply this decomposition in two steps. Table 4 documents how mean levels of entry and funding depth vary across income groups and decomposes the cross-country variance of ln(V /P) into contributions from each margin. We then examine country-level residuals on each margin relative to income predictions, which reveal systematic diferences in the structure of VC across country types.

High-income countries lead on both margins, but the gap is far wider in entry. The median high-income economy generates roughly 40 times more seed deals per capita than the median low- and lower-middleincome country and 11 times more than the median upper-middle-income country. Each entrant also attracts more funding, though the gap is narrower: \$7.0 million in high-income countries versus \$1.8 million in upper-middle-income and \$1.7 million in low- and lower-middle-income economies. Entry, not funding depth, is where income groups separate most, and these compounding diferences produce VC per capita ratios that difer by two orders of magnitude.

The role of entry is clearest within income groups. In high-income countries, entry is the separating margin: funding depth per entrant is relatively similar across these economies, since investors in most high-income markets can deploy comparable amounts of capital per firm, so what distinguishes Israel or Estonia from Japan or Italy is the rate at which new startups enter the VC pipeline (Panel B). Entry alone accounts for 57.1% of the within-group variance in high-income VC per capita, against 17.8% for funding depth. In developing countries the contrast is instructive: entry is uniformly low and has little room to vary. The median upper-middle-income country generates fewer than one seed deal per million people, and the median low-income country fewer than one-third of that. With low entry ratios, the margin that separates developing countries with active VC markets from those without is whether the investor base is deep enough to fund follow-on rounds. Funding depth accounts for 47.1% of within-group variance in low and lower-middle-income countries and 59.6% in upper-middle-income economies.

Table 4: Levels and Variance Decomposition of ln(VC/pop) across Countries (2021-2023)

	Full sample	HICs	UMICs	Low & LMICs
Panel A: Entry, funding, and VC intensity by income group
Entry: seed deals per million people (median)	1.09	10.54	0.96	0.26
Entry: seed deals per million people (mean)	7.96	20.68	1.22	0.60
Funding Depth: USD M per entrant (median)	3.3	7.0	1.8	1.7
Funding Depth: USD M per entrant (mean)	6.2	7.5	8.9	2.9
VC/pop: USD per capita (median)	3.9	74.2	1.9	0.4
VC/pop: USD per capita (mean)	67.5	180.6	7.9	1.6
Countries	148	53	41	54
Panel B: Variance decomposition of ln(VC/pop)
Var( ln (VC/pop.))	10.1	2.9	4.7	6.0
Var( ln(Seed deals/pop.))	4.3 (42.4%)	1.7 (57.1%)	1.3 (28.2%)	1.8 (29.5%)
Var( ln(Funding Depth))	2.5 (24.8%)	0.5 (17.8%)	2.8 (59.6%)	2.8 (47.1%)
2*Cov( ln(Seed deals/pop.),ln(Funding Depth))	3.3 (32.9%)	0.7 (25.1%)	0.6 (12.2%)	1.4 (23.4%)
Countries	148	53	41	54

Notes: Panel A reports the group median and mean of level variables. Entry is the number of seed deals per million inhabitants. Funding is the average total funding divided by the number of seed deals (USD millions per entrant). VC/pop is total VC divided by population (USD per capita). Panel B uses the exact identity V ar(ln V C/P) = Var(ln N₁/P) + Var(ln F D) + 2 Cov(ln N₁/pop, ln F D); parentheses show each component’s share of total variance (%). HICs = High income countries; UMICs = Upper-middle income countries; LMICs = Low-income and Lower-middle income countries. HICs includes USA; UMICs includes China; LMICs includes India. Income group classification based on World’s Bank 2025 classification.

As noted above, the additive identity lets us decompose a country’s deviation in VC depth (measures relative to population) into deviations on each margin²⁰.

(4)

For each country c, we regress each log margin on log GDP per capita and take the residual as its deviation from the level predicted by income. The residual decomposition is exact: a country’s over- or under-performance in VC depth, relative to its income, is the sum of its over- or under-performance in entry and in funding depth. Figure 7 shows each country’s position according to how its entry rate and funding depth deviate from what income alone would predict: the horizontal axis measures the residual on funding per entrant, the vertical axis the residual on entry per capita. Countries in the upper-right quadrant outperform on both margins; those in the lower-left underperform on both. Countries along the axes reveal which margin is farther from its expected value: a country in the lower-right has an investor base that funds startups well beyond what income would predict, but too few startups entering the VC pipeline to absorb that capital.

The figure reveals that even among countries that achieve similar overall VC outcomes, the underlying structure can difer. The United States and India both overperform their income-based expectations, but through diferent combinations: the US is strong on both margins, while India’s outperformance is driven more heavily by funding depth, consistent with a concentrated investor base funding a relatively narrow set of firms at scale. Estonia does so primarily through entry — a broad pipeline of startups per capita, while India’s strength lies on the funding side. At the other end, countries like Iraq underperform primarily on entry, while Paraguay and Angola fall short on both dimensions.

This heterogeneity has direct implications for policy. A country that underperforms on entry but meets or exceeds expectations on funding depth faces a diferent challenge than one where startups enter at expected rates but cannot attract follow-on capital. The first needs interventions that lower barriers to firm creation and experimentation. The second needs reforms that deepen the investor base, or measures that attract institutional and foreign capital. Our funding-depth measure does not capture these distributional efects but identifies countries where insuficient capital per entrant may be an important constraint.

Prior work establishes that institutional quality, investor protection, IPO activity, and tax policy shape investment and VC activity (La Porta et al., 1998; Gompers and Lerner, 1998; Cumming et al., 2010; Jeng and Wells, 2000). Our decomposition builds on this work to ask a more specific question: through which margin does each relationship run? Table 5 reports these regressions. Because the identity is additive in logs, the coeficient on VC depth in column 1 is the exact sum of the entry and funding-depth coeficients in columns 2 and 3, and column 4 reports the share that runs through entry. This split is exact, a property of the identity rather than of the estimates, so it attributes each established relationship to a margin.

For most variables, the association with VC depth runs through entry. Political stability loads almost entirely on this margin: 83% of its association with VC depth runs through the rate at which firms enter rather than the capital each raises, consistent with the idea that founding a firm requires a planning horizon and that entrepreneurs are less likely to enter where political disruption can invalidate business plans. This aligns with investor-side evidence: Colonnelli et al. (2025) find that investors penalize politica and country risk when evaluating deals, holding firm attributes fixed. Lower stability may also weaken demand, reducing the market potential that draws entrepreneurs in. The structure of taxation works the same way. For tax loss carryforward, 93% of the association with VC depth runs through entry: startups typically lose money in their early years, and a longer carryforward window reduces the expected cost of failure by allowing firms to ofset early losses against future profits, so where windows are shorter entrepreneurs face stronger incentives to pursue lower-risk models rather than the longer-horizon ventures VC typically finances.

Figure 7: Over/Under Performance Margin across Countries

Sources: Harmonized PitchBook and Crunchbase dataset; World Bank World Development Indicators.
Notes: The figure plots country-level residuals for two components of venture capital intensity over 2021–2023. The horizontal axis measures lifetime VC funding per entrant relative to the level predicted by GDP per capita, while the vertical axis measures VC entry per capita relative to the level predicted by GDP per capita. Dashed lines mark zero residuals. Countries to the right of the vertical line have larger average VC funding per entrant than expected given their income level, while countries above the horizontal line have higher VC entry per capita than expected. Countries are colored according to their overall VC-to-GDP performance relative to income expectations: underperformance is defined as less than −0.5 standard deviations, overperformance as greater than +0.5 standard deviations, and around expectation as within ±0.5 standard deviations.

The corporate income tax is the most revealing case, because the aggregate regression alone would miss it. Its association with VC depth is negative but indistinguishable from zero (column 1). The decomposition shows two ofsetting forces. The tax rate is strongly negatively associated with entry (−0.047, column 2) and positively, though imprecisely, associated with funding depth (+0.022, column 3), and the two cancel in the aggregate. Higher corporate taxes are associated with fewer firms entering the VC market, while the entrants that do appear may raise somewhat more funding on average, consistent with a more selected pool.

Not every factor concentrates on entry. Minority investor protection is associated with both margins at similar magnitudes, with roughly half the association running through each. When minority shareholders have legal recourse against expropriation, investors face lower risk and entrepreneurs can raise capita without ceding full control, consistent with the foundational evidence that investor protection is associated with financial development broadly (La Porta et al., 1997, 1998). Equity issuance activity sits at the other end: its association with VC depth runs predominantly through funding depth rather than entry, 72% through depth. Where public and private equity markets are deeper, the capital available to each entrant is larger.

It is important to note that the weaker associations with funding depth do not imply that the business environment is not important for how much capital firms raise. The depth margin is observed only for firms that have already entered, so it conditions on the selection that entry embodies. Business-environment conditions operate first at the entry gate; among the firms that clear it, much of that variation has already been absorbed, leaving a weaker association with the capital each goes on to raise. This selection also shapes which kinds of firms enter, and so what VC finances, a margin we examine by sector in the next table.

Table 5: Business Environment and VC Ecosystem Margins (2021–2023)

	(1)	(2)	(3)	(4) Ratio entry coef.
VARIABLES			ln(VC/pop) ln(Seed deals/pop) ln(Funding Depth)	(2)/(1)
Minority investor protection (0–100)	0.033*** (0.013)	0.017**	0.017**	0.502
WGI: Political stability (0–100)	0.048***	(0.006) 0.040***	(0.008) 0.008	0.832
Tax loss carryforward (years)	(0.009) 0.095***	(0.005) 0.088***	(0.006) 0.007	0.926
Effective capital gains tax rate (pct)	(0.024) 0.001	(0.015) 0.007	(0.015) -0.006
Corporate income tax rate (pct)	(0.024) -0.025	(0.013) -0.047***	(0.014) 0.022	1.881
Log equity issuance / GDP	(0.028) 0.088***	(0.016) 0.025*	(0.018) 0.063***	0.284
Constant	(0.025) -3.161** (1.355)	(0.013) -2.606*** (0.699)	(0.015) -0.555
Observations			(0.877)
R-squared	121 0.679	121 0.730	121 0.429

Notes: Minority investor protection and WGI political stability are min-max normalized to a 0–100 scale. Column (4) shows the ratio of the entry coeficient relative to the coeficient on ln(VC/pop). Ratio reported for variables with at least one significant coeficient. Ratios exceeding one arise when the two margins ofset and the total is close to zero. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

The associations examined so far pool all sectors, leaving open whether they operate uniformly across them. This matters because VC composition itself varies with development: Subsection 3.1 documented a tilt toward organizational-intensive sectors in developing economies, with knowledge-intensive sectors gaining share as VC markets deepen (Figure 5a). Table 6 re-estimates the regressions separately for entry into and funding depth within knowledge-intensive, organizational-intensive, and traditional sectors.

For most variables, the entry associations persist within sectors, at diferent magnitudes. Entry into knowledge-intensive sectors is the most sensitive to business-environment conditions: the coeficients on political stability, corporate income tax, and equity issuance are significantly larger in magnitude for knowledge-intensive entry than for traditional entry, and for corporate income tax and equity issuance the gap holds against organizational-intensive entry as well. Tax loss carryforward shows a similar but weaker pattern. These diferences are consistent with firms in knowledge-intensive sectors facing longer time-to-profit horizons and depending more on public equity exits, both of which leave them more exposed to institutional and policy conditions.

Table 6: Business Environment and VC Ecosystem Margins by Sector (2021–2023)

	(1)	(2)	(3)	(4)	(5)	(6)
	Knowledge-intensive		Organizational-intensive		Traditional
	ln(seed/pop)	ln(FD)	ln(seed/pop)	ln(FD)	ln(seed/pop)	ln(FD)

Minority investor protection (0–100)	0.010 (0.008)	0.000 (0.009)	0.007 (0.007)	0.008 (0.011)	0.009 (0.007)	0.008 (0.018)
WGI: Political stability (0–100)	0.048*** (0.006)	0.006 (0.007)	0.047*** (0.007)	-0.002 (0.008)	0.038*** (0.005)	-0.002 (0.010)
Tax loss carryforward (years)	0.098*** (0.017)	0.027 (0.020)	0.091*** (0.015)	0.029 (0.023)	0.088*** (0.013)	0.036 (0.029)
Effective capital gains tax rate (pct)	-0.001 (0.013)	-0.011 (0.016)	-0.008	-0.009	-0.010	-0.019
Corporate income tax rate (pct)	-0.064*** (0.018)	0.022	(0.011) -0.039**	(0.018) 0.012	(0.010) -0.025*	(0.018) -0.006
Log equity issuance / GDP	0.028* (0.016)	(0.022) 0.068***	(0.017) 0.008	(0.025) 0.101***	(0.014) 0.006	(0.032) 0.092***
Constant	-2.743*** (0.843)	(0.024) 0.427 (1.046)	(0.015) -3.901*** (0.887)	(0.023) 0.613 (1.286)	(0.015) -4.574*** (0.783)	(0.033) 1.351 (1.592)
Observations	94	94	94	94
R-squared	0.742	0.297	0.708	0.400	94 0.718	94 0.266

Notes: OLS regressions with robust standard errors. Each column restricts the dependent variable to deals in the indicated sector (K = knowledge-intensive; O = organizationally-intensive; T = traditional). Sample restricted to countries with non-missing values on all six outcomes and all six regressors in 2021–2023. Outcomes winsorized at top and bottom 1%. Minority investor protection and WGI political stability are min-max normalized to a 0–100 scale. Cross-sector equality of coeficients is tested via one-sided Wald tests on stacked equations using a robust joint covariance matrix, restricted to variables that are significant in the K columns. For political stability (entry), the coeficient for K and O are each significantly larger than that for T. For the corporate income tax rate (entry), the coeficient for K is significantly more negative than for O and for T, and the coeficient for O is significantly more negative than for T. For equity issuance (entry), the coeficient for K is significantly larger than for O and for T. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

Two of these diferential sensitivities themselves vary by country income. The association between equity issuance and entry into knowledge-intensive sectors is concentrated in high-income countries: in developing economies, equity market depth does not distinguish knowledge entry from other sectors, consistent with the IPO pathway operating primarily where public equity markets are mature. Corporate income tax sensitivity is present in both subsamples, with a sharper cross-sector gradient in high-income countries. Tax loss carryforward runs the other way, and is the most telling of the three: its association with knowledge-intensive entry appears in developing economies, where the coeficient is significantly larger for knowledge entry than for other sectors. Where other supporting institutions are weaker, loss-treatment provisions appear to do relatively more of the work in drawing knowledge-intensive startups into the market. Across all three, the channels linking the business environment to knowledge entry are themselves income-dependent, moving in the same direction as the composition gap documented in subsection 3.1.

These results show why the margins behind VC are worth separating. The shift in sectoral composition operates through how entry and funding depth respond to business conditions, and the decomposition lets us say how much each margin contributes. Entry accounts for the largest part of the cross-country gap in VC depth and carries most of its association with the business environment, and that association is most concentrated in entry into knowledge-intensive sectors. Because the conditions that favor knowledge entry are weaker in developing economies, knowledge entry recedes there, a pattern consistent with the sectoral tilt documented in subsection 3.1. Separating the margins thus shows that the business environment is linked to how much VC a country has and to what that VC finances primarily through the same margin, entry, and most through entry into the sectors that define the high-income model.

4 Conclusion

We constructed a comprehensive cross-country dataset on venture capital by harmonizing PitchBook and Crunchbase at the firm level and matching the result with LSEG equity issuance data at the sector level. This dataset enables analysis that moves beyond aggregate VC volumes to examine what VC finances across countries and how it compares to other forms of equity financing.

The evidence challenges the view that VC markets difer across countries only in scale. They difer in kind. VC is a major channel of equity finance in many developing economies, accounting for roughly two-thirds of equity financing in the average low- and lower-middle-income country. What it finances therefore matters for these economies. And it does not allocate capital the same way: rather than replicating the highincome pattern, VC in developing economies tilts toward organizational intangibles such as distribution, logistics, and payments, while high-income VC concentrates in knowledge intangibles tied to R&D and patenting. This diference is specific to VC. It has no counterpart in public equity issuance, where sectoral composition is far more similar across countries, which indicates that the tilt reflects something about how VC selects what to finance rather than about the underlying economies.

What VC finances is itself a marker of how developed a country’s VC market is, one that standard measures of financial development miss. To examine what lies behind these diferences, we decompose VC depth into the rate at which firms enter the market and the funding each entrant attracts. Entry accounts for the largest part of the cross-country gap and carries most of the association between VC and the business environment. That association is strongest for entry into knowledge-intensive sectors, the sectors most dependent on long horizons, R&D, and the protection of intangible assets. This links local conditions directly to what VC finances: where the conditions favorable to knowledge entry are weaker, knowledge entry recedes, a plausible channel for the shift toward organizational sectors observed across developing economies.

These findings have implications for how policymakers think about promoting VC in developing economies. The standard case for VC promotion rests on evidence that it finances knowledge-intensive innovation and generates technological spillovers (Kortum and Lerner, 2000; Popov and Roosenboom, 2012). That evi dence is drawn from settings where VC does finance knowledge-intensive activity. In developing economies, where VC finances organizational intangibles instead, the relevant spillovers are likely diferent: market creation, formalization, and financial inclusion rather than patenting and R&D. This does not diminish the case for VC promotion, but it changes what that case rests on. Policies designed to stimulate VC by replicating the institutional conditions that support knowledge-intensive entry may be less efective where the opportunities are organizational. And among developing economies, where entry rates are uniformly low, it is funding depth rather than entry that separates active VC markets from inactive ones. And the balance between the two margins itself varies widely across economies around the world, so the binding constraint, whether too few firms enter or too little capital is available to scale them, difers from country to country.

Important questions follow from these findings. Whether the composition of foreign and domestic in vestors helps explaining the knowledge–organizational split we document remains an open question. How exit channels shape investor behavior and sectoral allocation, and whether the returns to organizational innovation match those of knowledge-intensive VC in high-income settings, are also important direction for future research. What emerges clearly from our evidence is that venture capital is not a single institution deployed uniformly across economies. It adapts to the economic and business environments in which it operates, and understanding what it actually finances is a prerequisite for understanding what it can be expected to deliver.

References

Ahlstrom, D. and Bruton, G. D. (2006), ‘Venture capital in emerging economies: Networks and institutional change’, Entrepreneurship theory and practice 30(2), 299–320.

Aizenman, J. and Kendall, J. (2012), ‘The internationalization of venture capital’, Journal of Economic Studies 39(3), 488–511.

Avnimelech, G. and Teubal, M. (2006), ‘Creating venture capital industries that co-evolve with high tech: Insights from an extended industry life cycle perspective of the israeli experience’, Research Policy 35(10), 1477–1498.

Baumol, W. J. (1996), ‘Entrepreneurship: Productive, unproductive, and destructive’, Journal of business venturing 11(1), 3–22.

Beck, T., Levine, R. and Loayza, N. (2000), ‘Finance and the sources of growth’, Journal of financial economics 58(1-2), 261–300.

Black, B. S. and Gilson, R. J. (1998), ‘Venture capital and the structure of capital markets: Banks versus stock markets’, Journal of Financial Economics 47(3), 243–277.

Brynjolfsson, E. and Hitt, L. M. (2003), ‘Computing productivity: Firm-level evidence’, Review of Economics and Statistics 85(4), 793–808.

Colonnelli, E., Cruz, M., Lerner, J. and Pereira-L´opez, M. (2025), How do emerging markets investors make decisions? evidence from venture capital and private equity. SSRN Working Paper 5842742.

Colonnelli, E., Cruz, M., Pereira-Lopez, M., Porzio, T. and Zhao, C. (2026), ‘Startups in africa’, National Bureau of Economic Research, Working Paper 35261 .

Corrado, C., Haskel, J., Jona-Lasinio, C. and Iommi, M. (2022), ‘Intangible capital and modern economies’, Journal of Economic Perspectives 36(3), 3–28.

Corrado, C., Hulten, C. and Sichel, D. (2009), ‘Intangible capital and US economic growth’, Review of Income and Wealth 55(3), 661–685.

Crunchbase (2023), ‘Crunchbase’, https://www.crunchbase.com.

Cumming, D., Schmidt, D. and Walz, U. (2010), ‘Legality and venture capital governance around the world’, Journal of Business Venturing 25(1), 54–72.

Demmou, L., Franco, G. and Stefanescu, I. (2020), Productivity and finance: The intangible assets channel—a firm level analysis, Working paper, OECD Economics Department.

Durufl´e, G., Hellmann, T. F. and Wilson, K. E. (2017), ‘From start-up to scale-up: Examining public policies for the financing of high-growth ventures’, CEPR Discussion Paper No. DP12004 .

Estrin, S., Korosteleva, J. and Mickiewicz, T. (2013), ‘Which institutions encourage entrepreneurial growth aspirations?’, Journal of business venturing 28(4), 564–580.

Ewens, M., Peters, R. H. and Wang, S. (2025), ‘Measuring intangible capital with market prices’, Management Science 71(1), 407–427.

Gompers, P. and Lerner, J. (1998), ‘What drives venture capital fundraising?’, Brookings Papers on Economic Activity: Microeconomics pp. 149–204.

Gompers, P. and Lerner, J. (2001), ‘The venture capital revolution’, Journal of Economic Perspectives 15(2), 145–168.

Gornall, W. and Strebulaev, I. A. (2021), The economic impact of venture capital: Evidence from public companies. SSRN Working Paper 2681841.

Groh, A. and Wallmeroth, J. (2016), ‘Determinants of venture capital investments in emerging markets’, Emerging Markets Review 29, 104–132.

Grover, A., Medvedev, D. and Olafsen, E. (2019), High-growth firms: Facts, fiction, and policy options for emerging economies, World Bank Publications.

Hall, B. H. and Lerner, J. (2010), The financing of r&d and innovation, in ‘Handbook of the Economics of Innovation’, Vol. 1, Elsevier, pp. 609–639.

He, B., Mostrom, L. I. and Sufi, A. (2024), Investing in customer capital, Technical report, National Bureau of Economic Research.

Hellmann, T. and Puri, M. (2002), ‘Venture capital and the professionalization of start-up firms: Empirical evidence’, Journal of Finance 57(1), 169–197.

Hochberg, Y. V., Ljungqvist, A. and Lu, Y. (2007), ‘Whom you know matters: Venture capital networks and investment performance’, Journal of Finance 62(1), 251–301.

Imbens, G. W. and Wooldridge, J. M. (2009), ‘Recent developments in the econometrics of program evaluation’, Journal of Economic Literature 47(1), 5–86.

Jeng, L. A. and Wells, P. C. (2000), ‘The determinants of venture capital funding: Evidence across countries’, Journal of Corporate Finance 6(3), 241–289.

Kaplan, S. N. and Lerner, J. (2017), Venture capital data: Opportunities and challenges, in J. Haltiwanger, E. Hurst, J. Miranda and A. Schoar, eds, ‘Measuring Entrepreneurial Businesses: Current Knowledge and Challenges’, Vol. 75 of NBER Studies in Income and Wealth, University of Chicago Press, Chicago, pp. 413–431.

Khanna, T. and Palepu, K. (1997), ‘Why focused strategies may be wrong for emerging markets’, Harvard Business Review 75(4), 41–51.

Kortum, S. and Lerner, J. (2000), ‘Assessing the contribution of venture capital to innovation’, RAND Journal of Economics 31(4), 674–692.

La Porta, R., Lopez-de Silanes, F., Shleifer, A. and Vishny, R. W. (1997), ‘Legal determinants of external finance’, Journal of Finance 52(3), 1131–1150.

La Porta, R., Lopez-de Silanes, F., Shleifer, A. and Vishny, R. W. (1998), ‘Law and finance’, Journal of Political Economy 106(6), 1113–1155.

Lerner, J., Liu, J., Moscona, J. and Yang, D. Y. (2026), Appropriate entrepreneurship? the rise of China and the developing world, Working Paper 24-061, Harvard Business School. Revised January 2026.

Lerner, J. and Nanda, R. (2020), ‘Venture capital’s role in financing innovation: What we know and how much we still need to learn’, Journal of Economic Perspectives 34(3), 237–261.

Lev, B. and Radhakrishnan, S. (2005), The valuation of organization capital, in ‘Measuring capital in the new economy’, University of Chicago Press, pp. 73–110.

Levine, R. (1997), ‘Financial development and economic growth: views and agenda’, Journal of economic literature 35(2), 688–726.

Meh, C. A. and Schmukler, S. L., eds (2025), Financing firm growth: The role of capital markets in lowand middle-income countries, International Finance Corporation Research Series, World Bank.

Peters, R. H. and Taylor, L. A. (2017), ‘Intangible capital and the investment-q relation’, Journal of Financial Economics 123(2), 251–272.

PitchBook, a Morningstar company (2025), ‘PitchBook’, https://pitchbook.com/.

Popov, A. and Roosenboom, P. (2012), ‘Venture capital and patented innovation: Evidence from Europe’, Economic Policy 27(71), 447–482.

Rajan, R. and Zingales, L. (1998), ‘Financial dependence and growth’, The American Economic Review .

Samila, S. and Sorenson, O. (2011), ‘Venture capital, entrepreneurship, and economic growth’, Review of Economics and Statistics 93(1), 338–349.

Wasi, N. and Flaaen, A. (2015), ‘Record linkage using Stata: Preprocessing, linking, and reviewing utilities’, The Stata Journal 15(3), 672–697.

Webb, J. W., Khoury, T. A. and Hitt, M. A. (2020), ‘The influence of formal and informal institutional voids on entrepreneurship’, Entrepreneurship Theory and Practice 44(3), 504–526.

Appendix A: Data Harmonization Process

This appendix provides the full implementation details of the data harmonization process used to link firms across PitchBook and Crunchbase. As discussed in Section 2.2, the harmonization procedure combines website-based matching, firm-name-based matching, and firm-description verification, followed by additional checks of the top firms within each country. This appendix documents the pre-processing steps, score thresholds, verification procedures, and the evolution of the matched and unmatched samples across stages.

Stage 0: Pre-Harmonization and Standardization

Before the harmonization process begins, we implement several pre-processing steps to improve the likelihood of obtaining high-confidence matches across PitchBook and Crunchbase. First, firm websites are standardized by removing common prefixes (such as https:// and www.) and sufixes (such as .com, .org, and related extensions). In addition, we retain only the text appearing before the first dot in the website string. For example, a raw website such as https://venture-capital.com/ is standardized to venturecapital.

Firm names are also standardized using the Stata command stnd compname following guidelines from Wasi and Flaaen (2015). This procedure removes common legal sufixes (such as Inc., Ltd., and LLC), punctuation, spacing inconsistencies, and other formatting diferences, thereby isolating a cleaner core firm name for comparison across datasets. For example, a firm recorded as U.S. Venture-Capital Ltd. would be reduced to the standardized core name US VENTURE CAPITAL.

Match Scores and Classification Rules

For the website- and firm-name-based matching stages, we use the Stata command reclink2 to generate candidate matches. Match scores range from 0.80 to 1. We define a perfect match as a candidate pair with a score equal to 1, indicating that the underlying strings are identical in content and ordering. Perfect matches are sent to the last verification stage. We define a non-perfect match as a candidate pair with a score between 0.80 and less than 1. Non-perfect matches are not accepted automatically and are instead subjected to an additional verification step based on firms’ long descriptions.

Stage 1: Website-Based Matching

In Stage 1, we conduct website-based matching between PitchBook and Crunchbase using the standardized website variable. To improve precision and reduce false positives, matching is restricted to firms located in the same country. Candidate pairs are classified as either perfect or non-perfect matches. Table A.1 presents illustrative examples of website-based candidate matches across selected score thresholds.

Table A.1: Examples of Website-Based Matches by Score

Score Range	PitchBook firm	Crunchbase firm	Classification	Later Matched
0.80-0.84	ariadneinternational	roa-international	Non-perfect match	No
0.85-0.89	rayalliance	dcalliance	Non-perfect match	No
0.90-0.95	mosh	moshimo	Non-perfect match	No
0.95-0.99	ladder	myladder	Non-perfect match	Yes
1.00	insightrobotics	insightrobotics	Perfect match

Notes: This table provides illustrative examples of website-based candidate matches between PitchBook and Crunchbase at diferent levels of the website-match score. The score reflects the similarity of website names after preprocessing and standardization. A score of 1.00 indicates a perfect website match, while scores above 0.80 and below 1.00 indicate non-perfect matches that proceed to additional verification in later stages of the harmonization process. The column Later Matched shows whether a non-perfect match was ultimately confirmed as a match in Stages 2–4, and then marked as “Yes”; entries marked “No” were not confirmed and therefore remained unmatched. For perfect matches, this column is reported as N/A because no further verification is needed.

Stage 2: Verification of Non-Perfect Website Matches

In Stage 2, firms that receive a website match score between 0.80 and less than 1 in Stage 1 are subjected to an additional verification step based on firm long descriptions. Specifically, we use Claude AI to compare the firm descriptions for each candidate pair and determine whether the two records refer to the same underlying firm. The prompt used for this exercise is provided below. This verification step yields one of two outcomes: match or no match. Firms classified as matches are retained as confirmed links and proceed to Stage 5, while firms classified as non-matches proceed to Stage 3 to continue through the matching process.

Summary of the Claude AI verification prompt

For candidate pairs with non-perfect website matches, the Claude AI prompt applies the following decision rules:

• Objective: Compare two startup records and determine whether they refer to the same underlying firm.

• Primary evidence: Base the decision primarily on the firm names and long descriptions provided in the matched records.

• Priority rule: Give greater weight to similarity in business descriptions than to similarity in firm names.

• Name handling: Ignore legal sufixes such as Inc., LLC, Corp., and Ltd.; allow for minor spelling diferences, abbreviations, and common naming variations.

• External knowledge: Use external knowledge, such as well-known aliases, abbreviations, or brand names, only when it is highly reliable.

• Decision logic: Classify a pair as a match when the descriptions clearly indicate the same firm, product, or service; classify it as a non-match when the descriptions point to diferent industries or oferings.

• Output format: Return a structured response containing a match classification, a confidence score, and a brief rationale.

Table A.2 summarizes the outcome of this verification step.

Table A.2: Stage 2 Results

Outcome	Share (%)
Verified matches	6.8
No-match decisions	93.2

Notes: This table reports the results of Stage 2, in which candidate pairs with non-perfect website matches from Stage 1 are evaluated using firm long descriptions. Claude AI is used to assess whether each pair refers to the same underlying firm. Verified matches proceed to Stage 5, while no-match decisions are not confirmed at this stage and proceed to Stage 3. Shares are reported as percentages of all observations entering Stage 2.

Stage 3: Firm-Name-Based Matching

In Stage 3, we repeat the matching procedure described in Stage 1, using standardized firm names instead of website URLs. The sample entering this stage consists of (1) firms classified as no match in Stage 2 and (2) firms that remained unmatched in Stage 1. As in Stage 1, the procedure yields three possible outcomes: unmatched firms, perfect matches that proceed to Stage 5, and non-perfect matches that advance to Stage 4 for additional verification. Table A.3 presents illustrative firm-name candidate matches across selected score thresholds.

Table A.3: Examples of Firm-Name Matches by Score

Score Range	PitchBook firm	Crunchbase firm	Classification	Later Matched
0.80-0.84	GWC ITALIA	SIF ITALIA	Non-perfect match	No
0.85-0.89	GRUPO NERVION	GRUPO CDV	Non-perfect match	No
0.90-0.95	BIOMATZ	BIOME ATS	Non-perfect match	Yes
0.95-0.99	HEALTHFLEX	EHEALTHFLEX	Non-perfect match	No
1.00	GMOBILE	GMOBILE	Perfect match

Notes: This table reports illustrative examples of firm-name-based matches between PitchBook and Crunchbase at diferent levels of the firm-name match score. The score reflects the similarity of firm names after preprocessing and standardization. Observations with a score of 1.00 are classified as perfect matches, while those with scores from 0.80 to less than 1.00 are classified as non-perfect matches and proceed to additional verification in subsequent stages of the harmonization process. The column Later Matched shows whether a non-perfect match was later confirmed as a match (marked as “Yes”) in Stage 4; entries marked “No” remained unmatched.

Stage 4: Verification of Non-Perfect Firm-Name Matches

In Stage 4, we replicate the verification procedure described in Stage 2 for the non-perfect firm-name matches generated in Stage 3. Using firm long descriptions, Claude AI evaluates whether each candidate pair corresponds to the same underlying firm. This stage yields two possible outcomes: match and no match. Confirmed matches are retained in the harmonized dataset, while candidate pairs classified as non-matches are carried forward as unmatched observations in the final dataset.

Stage 5: Validation of Top-Raising Firms

In Stage 5, we conduct an additional validation exercise for the top 60 matched firms in each country dataset with the highest total amount of venture capital received. The validation sample consists of perfect matches identified in Stages 1 and 3 and non-perfect matches classified as match in Stages 2 and 4. We use Claude AI to compare the firm descriptions associated with each matched pair and determine whether the two records refer to the same underlying firm. Matches confirmed through this review are retained, while pairs classified as non-matches are returned to the unmatched pool. This stage is designed to reduce the risk of false positives among firms accounting for the largest amounts of venture capital activity.

Final Consolidation

After incorporating the results of the Stage 5 validation exercise, we consolidate the outputs of the harmonization process into matched and unmatched firm datasets. The matched-firm dataset includes perfect website and firm-name matches, non-perfect matches confirmed through description-based verification, and matches retained following the Stage 5 review. The unmatched-firm dataset includes firms that were not successfully linked during Stages 1 through 4, candidate pairs rejected during description-based verification, and matches rejected during the Stage 5 validation exercise. Table A.4 shows that fuzzy matching below a score of 1.0 recovers genuine matches that would otherwise be missed because of minor diferences in URLs or firm names.

Table A.4: Description Verification Outcomes by Score Bin

Score Range	Probability of Confirmed Matches (%)
0.80-0.84	2.0
0.85-0.89	3.1
0.90–0.94	2.4
0.95-0.99	8.3
1.00	97.8

Source: Harmonized PitchBook and Crunchbase dataset.
Notes: This table shows the probability that a website-based match is confirmed by description verification across bins of the websitematch score. Description verification is used as an additional check on whether matched PitchBook and Crunchbase records refer to the same firm. Probabilities are reported as percentages.

Appendix B: Data

1. Venture Capital

This paper uses PitchBook and Crunchbase to construct the harmonized venture capital dataset employed in the empirical analysis. As discussed in Section 2.1, these sources provide the core firm- and deallevel information used throughout the paper. This appendix first documents the underlying sample in PitchBook and Crunchbase and then compares the composition of the two sources.

Table B.1 summarizes the raw coverage of each source. Tables B.2 and B.3 report sample size by income group and by region, respectively. The statistics presented in these tables are based on the full underlying data, with no sample restrictions applied: they include all firms regardless of deal type and cover firms with observed deals during 1990–2023. These tables provide a benchmark for the raw coverage of each source across countries and regions.

Table B.5 presents summary statistics of the sample used in the paper. Table B.4 then examines how the composition of the two sources difers among unique, non-matched deals. Specifically, it compares mean shares across Crunchbase and PitchBook by sector, ticket size, income group, and region, and reports tests of statistical diference and normalized diferences. This comparison is useful for assessing whether the gains from harmonization reflect only a larger sample or also a change in the composition of the observed venture capital universe.

Overall, the tables in this appendix show that PitchBook and Crunchbase difer along several important dimensions, reinforcing the value of combining the two sources in a harmonized dataset.

Table B.1: Summary of PitchBook and Crunchbase Data

Stage	PitchBook	Share of PitchBook (%)	Crunchbase	Share of Crunchbase (%)
Total Firms	520,257	100.0	201,589	100.0
VC Firm	253,768	48.8	183,977	91.3
VC + Young Firm	227,734	43.8	169,322	84.0

Sources: PitchBook, a Morningstar company, and Crunchbase.

Notes: This table reports the sample size in PitchBook and Crunchbase across key stages of sample construction. “Total Firms” includes all firms in each source regardless of deal type, “VC Firm” restricts the sample to firms with venture capital or seed financing, and “VC + Young Firm” further restricts the sample to VC-backed young firms, defined as firms that received venture capital or seed financing before age 15. Shares are expressed relative to the total number of firms in each source.

Table B.2: Sample Size by Income Group
Panel A. PitchBook

	Total	High-income	Upper-middle	Lower-middle	Low-income
Number of Firms	520,257	448,562	46,755	24,006	862
Number of Countries	200	77	51	49	22

Panel B. Crunchbase

	Total	High-income	Upper-middle	Lower-middle	Low-income
Number of Firms	201,589	167,202	21,960	12,149	255
Number of Countries	167	64	45	39	18

Sources: PitchBook, a Morningstar company, and Crunchbase
Notes: This table reports the sample size of firms by income group in PitchBook (Panel A) and Crunchbase (Panel B). Country income group classifications follow the World Bank’s 2025 classifications.

Table B.3: Sample Size by Region
Panel A. PitchBook

	Number of Firms	Number of Countries
East Asia and Pacific	70,466	31
Europe and Central Asia	131,483	55
Latin America and Caribbean	12,992	38
Middle East and North Africa	9,782	21
North America	272,329	3
South Asia	17,077	8
Sub-Saharan Africa	6,128	44

Panel B. Crunchbase

	Number of Firms	Number of Countries
East Asia and Pacific	27,386	22
Europe and Central Asia	49,882	51
Latin America and Caribbean	5,565	29
Middle East and North Africa	5,327	19
North America	102,110	3
South Asia	8,888	6
Sub-Saharan Africa	2,431	37

Sources: PitchBook, a Morningstar company, and Crunchbase
Notes: This table reports the sample size of firms by region in PitchBook (Panel A) and Crunchbase (Panel B). Regional group classifications follow the World Bank’s 2025 classifications.

Table B.4: Diferences Between Unique Deals by Data Source

	Mean Unique CB	Mean Unique PB	P-value	Norm. Diff.
Sector
IT	0.459	0.401	0.000	0.083
Manuf. - Chemicals, Pharma	0.022	0.059	0.000	-0.133†
Manuf. - Electronics	0.089	0.156	0.000	-0.147†
Manuf. - Food and Bev.	0.015	0.033	0.000	-0.083
Manuf. - Others	0.025	0.055	0.000	-0.107†
Financial Services	0.051	0.030	0.000	0.074
Wholesale and Retail	0.064	0.036	0.000	0.089
Services - R & D	0.049	0.048	0.238	0.003
Services - Health	0.036	0.028	0.000	0.032
Telecommunications	0.008	0.006	0.000	0.020
Agriculture and Natural Resources	0.021	0.028	0.000	-0.034
Utilities, Energy and Waste	0.009	0.017	0.000	-0.053
Construction	0.003	0.005	0.000	-0.016
Transportation and Logistics	0.020	0.012	0.000	0.042
Tourism and Leisure	0.034	0.025	0.000	0.035
Real Estate and Prof. Services	0.068	0.043	0.000	0.078
Education	0.029	0.017	0.000	0.055
Ticket Size
Ticket size: <=$100k	0.191	0.238	0.000	-0.082
Ticket size: $100k-$3M	0.573	0.529	0.000	0.062
Ticket size: $3M+	0.236	0.232	0.021	0.006
Country income
Low and lower-middle income	0.065	0.053	0.000	0.035
Upper-middle income	0.189	0.106	0.000	0.168†
High income	0.746	0.841	0.000	-0.168†
Region
East Asia Pacific	0.212	0.161	0.000	0.093
Europe and Central Asia	0.218	0.298	0.000	-0.131†
Latin America and Caribbean	0.041	0.019	0.000	0.090
Middle East and North Africa	0.028	0.022	0.000	0.030
North America	0.442	0.451	0.000	-0.012
South Asia	0.044	0.038	0.000	0.021
Sub-Saharan Africa	0.015	0.011	0.000	0.026
Observations	94,304	216,837

Source: Harmonized PitchBook and Crunchbase dataset.
Notes: This table reports diferences in the composition of unique, non-matched deals across Crunchbase and PitchBook. Entries in the first two columns are mean shares by data source. The p-value column reports the result of a t-test for equality in means, and the normalized diference column reports the diference in means scaled by the pooled standard deviation. Normalized diferences are reported in absolute value. Following Imbens and Wooldridge (2009), |ND| > 0.1 suggests potential imbalance between groups. Sectoral groupings are based on ISIC classifications, and country income group and regional classifications follow the World Bank’s 2025 classifications. The t-test is conducted between non-matched firms in Crunchbase and PitchBook.

Table B.5: Descriptive Statistics at Deal and Firm-Level

	All	High Income	Developing Countries (excl. China)	China
Panel A: Deal size
Ticket size: <=$100k	19.1%	19.4%	31.5%	1.2%
Ticket size: $100k-$3M	54.2%	55.1%	51.3%	47.6%
Ticket size: $3M+	26.7%	25.5%	17.2%	51.2%
Panel B: Sector (firm-level)
IT	46.3%	47.8%	45.3%	36.1%
Manuf. - Chemicals, Pharma	4.0%	4.1%	2.0%	5.0%
Manuf. - Electronics	11.3%	11.3%	5.1%	19.2%
Manuf. - Food and Bev.	2.9%	3.0%	2.7%	2.0%
Manuf. - Others	4.2%	4.3%	2.9%	4.8%
Financial Services	3.5%	3.2%	6.2%	2.7%
Wholesale and Retail	4.8%	4.0%	8.6%	7.5%
Services - R & D	4.2%	4.7%	1.2%	3.4%
Services - Health	3.1%	3.0%	3.6%	2.5%
Telecommunications Agriculture and Natural Resources	0.3%	0.3%	0.4%	0.3%
	2.4%	2.3%	4.3%	1.2%
Utilities, Energy and Waste Construction	1.3% 0.4%	1.3%	1.8%	0.9%
Transportation and Logistics	1.6%	0.4% 1.2%	0.5% 3.7%	0.4%
Tourism and Leisure	2.8%	2.7%	3.2%	2.1%
Real Estate and Prof. Services				3.1%
	4.9%	4.8%	5.6%	5.0%
Education	2.0%	1.6%	3.1%	3.8%
Panel C: Investment (firm-level)
Prob. of receiving follow-on investment	43.9%	45.9%	36.7%	35.5%
Panel D: Region (firm-level) East Asia and Pacific
Europe and Central Asia	18.0% 27.4%	9.4% 33.3%	11.1% 7.4%	100.0%
Latin America and Caribbean	2.7%	0.7%	21.0%	0.0%
Middle East and North Africa	2.7%	2.7%	5.3%	0.0%
North America			0.0%	0.0%
South Asia	43.2%	53.8%	42.7%	0.0%
Sub-Saharan Africa	4.5%	0.0% 0.0%	12.5%	0.0% 0.0%
	1.3%
Panel E: Income group (firm-level)
Low and lower-middle income	6.5%	0.0%	61.1%	0.0%
Upper-middle income	13.6%	0.0%	38.9%	100.0%
High income	79.9%	100.0%	0.0%	0.0%
N deals	527,031	436,273	49,338	41,420
N firms	269,915	215,729	28,731	25,455

Source: Harmonized PitchBook and Crunchbase dataset
Notes: This table reports descriptive statistics for the sample used in the analysis at the deal and firm levels, separately for the full sample, high-income countries, non-high-income countries, and China. Panel A shows the distribution of deal size categories. Panels B–E report firm-level shares by sector, the probability of receiving a follow-on investment, region, and income group. Sectoral groupings are based on ISIC classifications. Country income group and regional classifications follow the World Bank’s 2025 classifications.

2. Equity Issuance

The raw equity issuance records come from the Securities Data Company (SDC) Platinum database provided by the London Stock Exchange Group (LSEG). SDC Platinum is a comprehensive transaction level database covering new bond issues, mergers and acquisitions, syndicated loans, and equity issuances. This paper uses the equity issuance component, which contains detailed information on transactions by publicly listed and privately held firms across a broad set of countries. Coverage begins in the 1970s for the United States and generally becomes more comprehensive for other markets from 1990 onward. The database is updated continuously, so transactions that are not captured initially may be incorporated in subsequent releases. Since each observation includes the issuance date, transactions can still be assigned to the period in which they occurred.

For this paper, we retain corporate equity issuances relevant to the comparison with venture capital and exclude observations outside the firm and security categories used in the analysis. The raw transaction level records are then cleaned and aggregated to the country-sector-year level. Sector classifications are standardized and mapped to the same ISIC-based categories used in the harmonized VC dataset, allowing the two sources to be compared consistently across countries and sectors.

Following the cumulative issuance approach in Meh and Schmukler (2025), we construct the equity issuance measure from a fixed baseline year. Whereas Meh and Schmukler (2025) use 1990 as the baseline, we use 2010 to align the equity issuance series with the period covered by the harmonized VC analysis. We therefore set all pre-2010 issuance to zero and measure equity issuance cumulatively from 2010 onward. This construction allows us to benchmark the scale and sectoral composition of VC activity against public equity markets.

Appendix C: Country and Economy Income Classifications

This section presents the countries, economies, special regions, and territories included in the report, classified by income group. To ensure that the number of countries in each group remains constant throughout the analysis, the report adopts a static income classification based on the World Bank’s 2025 income categories. Using a fixed classification avoids changes in group composition over time, which could otherwise make it dificult to distinguish whether observed trends reflect changes in capital market activity or changes in countries’ income status. The World Bank assigns economies to income groups based on gross national income (GNI) per capita, expressed in U.S. dollars. In this report, countries are grouped into four categories: high-income, upper-middle-income, lower-middle-income, and low-income economies. The tables below list the economies included in each income group.

Table C.1: High and Upper-Middle Income Countries and Economies Examined

High	High cont.	Upper-middle	Upper-middle cont.
Australia	Lithuania	Albania	Iraq
Austria	Luxembourg	Algeria	Jamaica
Bahrain	Macao SAR, China	Argentina	Kazakhstan
Belgium	Malta	Armenia	Kosovo
Brunei	Netherlands	Azerbaijan	Libya
Bulgaria	New Zealand	Belarus	Malaysia
Canada	Norway	Bosnia and Herzegovina	Mauritius
Chile	Oman	Botswana	Mexico
Croatia	Panama	Brazil	Moldova
Cyprus	Poland	China	Mongolia
Czech Republic	Portugal	Colombia	Montenegro
Denmark	Qatar	Costa Rica	Namibia
Estonia	Romania	Cuba	North Macedonia
Finland	Russian Federation	Dominican Republic	Paraguay
France	Saudi Arabia	Ecuador	Peru
Germany	Singapore	El Salvador	Serbia
Greece	Slovak Republic	Fiji	South Africa
Guyana	Slovenia	Gabon	Thailand
Hong Kong SAR, China	Korea, Rep.	Georgia	Türkiye
Hungary	Spain	Guatemala	Ukraine
Ireland	Sweden	Indonesia
Israel	Switzerland	Iran, Islamic Rep.
Italy	Trinidad and Tobago
Japan	United Arab Emirates
Kuwait	United Kingdom
Latvia	United States
	Uruguay

Source: The classification year of countries into income groups is 2025 using the World Bank income categories.
Note: The table presents the countries, economies, special regions, and territories included in this paper, classified as high-income or upper-middle-income.

Table C.2: Lower-Middle and Low-Income Countries and Economies Examined

Lower-middle	Lower-middle cont.	Lower-middle cont.	Low	Low cont.
Angola Bangladesh Benin Bolivia Cambodia Cameroon Cape Verde Congo, Rep. Egypt, Arab Rep. Ghana Guinea Haiti Honduras	India Ivory Coast Jordan Kenya Kyrgyz Republic Lao PDR Lebanon Lesotho Mauritania Morocco Myanmar	Nepal Nicaragua Nigeria Pakistan West Bank and Gaza Philippines Senegal Sri Lanka Swaziland Tajikistan Tanzania Timor-Leste Tunisia Uzbekistan Viet Nam	Afghanistan Burkina Faso Burundi Chad Ethiopia Liberia Madagascar Malawi Mali Mozambique	Niger Rwanda Sierra Leone Somalia Sudan Gambia, The Togo Uganda Yemen, Rep.

Source: The classification year of countries into income groups is 2025 using the World Bank income categories.

Appendix D: Sector Classifications

This section describes the classifications used to distinguish between knowledge-intensive, organizationalintensive, and traditional venture capital activity. The primary classification is based on sectors. Each deal in the harmonized venture capital dataset is first classified into one of 17 sectors based on the firm’s ISIC classification. The underlying 2-digit ISIC division is then used to assign sectors to broader groups according to their intangible intensity. This classification builds on Demmou et al. (2020), which trans lates the firm-level intangible capital framework of Peters and Taylor (2017) into sector-level measures. In this framework, intangible capital includes knowledge capital, measured through accumulated R&D spending, and organization capital, captured through selling, general, and administrative expenses associated with management practices, brands, distribution systems, and human capital investments. Most knowledge-intangible sectors are also classified as organizationally intensive in the underlying intangibleintensity framework, with the exception of Manufacture of basic pharmaceuticals and pharmaceutica preparations. To present mutually exclusive categories, overlapping sectors are assigned first to the knowledge-intangible group. The organizational-intangible category therefore refers to sectors that are organizationally intensive but not knowledge-intensive, while traditional sectors are those classified as neither knowledge- nor organizational-intangible. Although these classifications are largely derived from firm-level data in high-income countries, they are useful for cross-country analysis because intangible intensity reflects the production technology of a sector rather than only the country in which firms operate.

We also complement the ISIC-based sector classification with a keyword-based classification using firm business descriptions. This approach captures knowledge- and organizational-intensive business-model characteristics that may not be fully reflected in sector codes alone. Firm descriptions are converted to lowercase, and descriptions with fewer than 10 words or boilerplate language such as ‘operating in stealth mode” or ‘undisclosed products” are excluded. A firm is classified as knowledge-intensive if its description contains at least one knowledge keyword, and as organizational-intensive if its description contains at least one organizational keyword.

The tables below list the 2-digit ISIC division classifications used in the report to group sectors into knowledge-intangible, organizational-intangible, and traditional activities, as well as the business-description keywords used to classify firms by knowledge- and organizational-intensive business models.

Table D.1: Knowledge-Intangible ISIC Division Classifications

ISIC Sector

Computer programming, consultancy and related activities Information service activities

Manufacture of basic pharmaceuticals and pharmaceutical preparations Manufacture of computer, electronic and optical products

Manufacture of electrical machinery and apparatus Manufacture of furniture

Manufacture of machinery and equipment n.e.c.

Other manufacturing Repair and installation of machinery and equipment

Notes: Sectors are classified using 2-digit ISIC divisions. This table lists sectors classified as knowledge-intensive. Most knowledge-intangible sectors are also organizationally intensive and are assigned to the knowledgeintangible category to keep groups mutually exclusive. The exception is Manufacture of basic pharmaceuticals and pharmaceutical preparations, which is classified as knowledge-intensive but not organizationally intensive.

Table D.2: Organizational-Intangible ISIC Division Classifications

ISIC Sector

Activities of head offices; management consultancy activities Advertising and market research

Architectural and engineering activities; technical testing and analysis Civil engineering

Construction of buildings

Employment activities

Legal and accounting activities Manufacture of beverages

Manufacture of fabricated metal products, except machinery and equipment

Manufacture of food products Manufacture of leather and related products

Manufacture of rubber and plastic products

Manufacture of textiles

Manufacture of tobacco products Manufacture of wearing apparel Office administrative, office support and other business support activities

Notes: Sectors are classified using 2-digit ISIC divisions. This table lists sectors classified as organizationally intensive but not knowledge-intensive. Most knowledge-intangible sectors are also organizationally intensive and are therefore assigned to the knowledge-intangible category to keep groups mutually exclusive. The exception is Manufacture of basic pharmaceuticals and pharmaceutical preparations, which is classified as knowledge-intensive but not organizationally intensive.

Table D.3: Traditional ISIC Division Classifications

ISIC Sector

Accommodation Air transport Crop and animal production, hunting and related service activities

Extraction of crude petroleum and natural gas Fishing and aquaculture

Food and beverage service activities Forestry and logging

Land transport and transport via pipelines

Manufacture of basic metals Manufacture of coke and refined petroleum products

Manufacture of motor vehicles, trailers and semi-trailers Manufacture of other non-metallic mineral products

Notes: Sectors are classified using 2-digit ISIC divisions. This table lists sectors classified as neither knowledge-intensive nor organizationally intensive. To keep the groups mutually exclusive, sectors that are both knowledge- and organizationally intensive are assigned to the knowledge-intangible category.

Table D.4: Knowledge- and Organizational-Intensive Business Model Keywords

Knowledge-intensive keywords Organizational-intensive keywords

proprietary; patent; patented; algorithm; machine distribution; logistics; last-mile; fulfillment; ware- learning; deep learning; artificial intelligence; r&d; house; supply chain; marketplace; aggregator; field research and development; clinical trial; drug dis- sales; delivery network; retail network; sales force; covery; biotech; semiconductor; chip design; propri- customer acquisition; offline-to-online; o2o; hy- etary technology; proprietary algorithm; deep tech; perlocal; kirana; dukan; agent network; branch computer vision; nlp; genomic; crispr; therapeutic; network; driver network; rider; courier; fleet; preclinical; scientific discovery; machine vision; nat- e-commerce; ecommerce; food delivery; mobile ural language; gene therapy; cell therapy; intellec- money; digital payment; ride hailing; ride sharing; tual property; neural network point of sale

Notes: A firm is classified as knowledge-intensive or organizational-intensive if its description contains at least one keyword from the corresponding list.

Appendix E: Additional Graphs

This appendix presents additional figures that support the main analysis. Figure E.1 shows the relationship between income levels and the relative size of venture capital and equity issuance activity, respectively, measured as a share of GDP. Figure E.3 complements the sectoral-composition analysis by comparing the distribution of cosine similarity scores for venture capital and equity issuance markets in developing economies. Figures E.4 and E.5 further examine whether sectoral similarity to the United States is related to countries’ income levels, measured by GDP per capita. Together, these figures provide additional evidence on the scale of VC activity across countries and on the extent to which VC sectoral composition difers from public equity markets.

Figure E.1: Average Size of VC relative to GDP

Sources: Harmonized PitchBook and Crunchbase dataset and GDP data from the World Bank’s World Development Indicators.

Figure E.2: Share of VC relative to Total Equity Financing

Sources: Harmonized PitchBook and Crunchbase dataset and equity data from calculations using issuance data from the Securities Data Company Platinum database from LSEG.

Figure E.3: Cosine Similarity to US for VC and Equity markets in developing countries

Sources: Harmonized PitchBook and Crunchbase dataset and equity data from calculations using issuance data from the Securities Data Company Platinum database from LSEG.
Note: This figure plots the kernel density of the cosine similarity index between each developing country’s sectoral distribution and the corresponding sectoral distribution in the United States, shown separately for venture capital and equity markets. The cosine similarity index ranges from 0 to 1, with higher values indicating greater similarity to the U.S. sectoral mix. The dark curve corresponds to VC and the light curve to equity markets. Vertical dashed lines mark the median of each distribution. Sectoral groups are based on ISIC classifications.

Figure E.4: Cosine Similarity to US for VC and GDP per Capita

Sources: Harmonized PitchBook and Crunchbase dataset and equity data from calculations using issuance data from the Securities Data Company Platinum database from LSEG and GDP per capita from World Bank’s World Development Indicators. Note: This figure plots the relationship between GDP per capita and the cosine similarity index between each country’s venture capital sectoral distribution and the corresponding sectoral distribution in the United States. The cosine similarity index ranges from 0 to 1, with higher values indicating greater similarity to the U.S. VC sectoral mix. The red line shows the fitted linear relationship. GDP per capita is measured as the log average over 2021–2023. The United States is excluded because it is the reference country. Sectoral groups are based on ISIC classifications.

Figure E.5: Cosine Similarity to US for Equity Markets and GDP per Capita

Sources: Equity data from calculations using issuance data from the Securities Data Company Platinum database from LSEG and GDP per capita from World Bank’s World Development Indicators.
Note: This figure plots the relationship between GDP per capita and the cosine similarity index between each country’s equity market sectoral distribution and the corresponding sectoral distribution in the United States. The cosine similarity index ranges from 0 to 1, with higher values indicating greater similarity to the U.S. equity sectoral mix. The red line shows the fitted linear relationship. GDP per capita is measured as the log average over 2021–2023. The United States is excluded because it is the reference country. Sectoral groups are based on ISIC classifications.

Appendix F: Robustness Checks

This appendix presents a series of robustness checks that complement the main results reported in the paper. The exercises assess whether the key findings on the sectoral composition of venture capital, intangible intensity, firm-level business models, and the entry and funding-depth margins remain robust to alternative outcome measures, sample definitions, weighting schemes, and variable constructions. To gether, these checks provide additional support for the patterns documented in the main text.

Table F.1: Distribution of VC across Sectors and Income Groups

	(1)	(2)	(3)	(4) High income	(5) Developing Countries
	United States	China India		(excl. United States)	(excl. China and India)
IT	43.2	28.0	36.2	39.4	30.2
Wholesale and Retail	3.3	10.1	21.7	7.3	17.7
Financial Services	5.9	3.6	8.9	8.6	15.1
Transportation & Logistics	2.2	4.9	8.1	3.4	11.2
Real Estate & Prof. Services	2.9	3.9	3.1	4.8	5.6
Tourism & Leisure	1.4	2.3	4.5	2.4	3.6
Utilities, Energy & Waste	1.1	1.2	1.1	2.1	3.4
Agriculture & Natural Resources Manuf. - Electronics	1.7	0.9 19.6	1.5	2.6	2.9
Manuf. - Others	11.1 4.0	8.0	1.8 3.0	10.6	1.9
Education	0.7	1.8	2.7	3.5	1.9
Services - Health	4.0	2.1	2.5	0.7 1.7	1.7
Manuf. - Chemicals, Pharma	5.5	7.0	1.5	4.8	1.7
Telecommunications	0.4	0.2	0.4	0.7	0.9
Manuf. - Food & Bev.	1.6	1.0	1.8	1.2	0.7
Construction	0.5	0.3	0.7	0.6	0.6
Services - R&D	10.5	5.1	0.5	5.7	0.6
N countries	1	1	1	53	0.4 105

Sources: Harmonized PitchBook and Crunchbase dataset and GDP data from World Bank’s World Development Indicators. Notes: This table reports the average share of VC volume accounted for by each sector over 2012–2023, within countries, for five groups: the United States, China, India, high-income countries excluding the United States, and low- and middle-income countries excluding China and India. Averages within the two multi-country groups are weighted by country GDP; the three single-country columns report that country’s own sectoral shares. Sectoral groupings are based on ISIC classifications, and income classifications are based on the World Bank’s 2025 income classifications. Values are expressed as percentages of total VC volume within countries.

Table F.2: Intangible Intensity and VC Deal Count Shares

	(1) Total	(2) Total x Inc.	(3)	(4) Knowledge Knowledge x Inc. Organizational Organizational x Inc.	(5)	(6)
High intangible intensity (total)	0.643*** (0.012)	0.680* (0.006) -0.203*
× Upper-middle income		(0.022) -0.081***
High intangible intensity (knowledge)		(0.016)	0.306*** (0.023)	0.364*** (0.009)
× Low & lower-middle income				-0.251*** (0.031)
× Upper-middle income				-0.144*** (0.020)
High intangible intensity (organizational)					0.666*** (0.014)	0.700*** (0.005)
× Low & lower-middle income						-0.158*** (0.024)
× Upper-middle income						-0.082*** (0.012)
Constant	0.111*** (0.005)	0.100*** (0.003)	0.280*** (0.011)	0.257*** (0.006)	0.100*** (0.007)	0.089*** (0.002)
Observations	483	483	483	483	483	483
R-squared	0.971	0.979	0.901	0.936	0.979	0.986
p-value: Low & LMI = 0		0.000		0.000		0.000
p-value: Upper-middle = 0		0.000		0.000		0.000

Notes: This table reports regressions of the share of VC deal counts across sectors on measures of sectoral intangible intensity. The dependent variable is the share of VC deals in each sector-country combination. Sectors are classified as meeting the relevant intangible-intensity definition, not meeting it, or lacking suficient information; regressions use the first two groups. Columns (1)–(2), (3)–(4), and (5)–(6) use the overall, knowledge, and organizational intangible intensity measures, respectively. Even-numbered columns interact intangible intensity with indicators for low- and lower-middleincome and upper-middle-income countries, with high-income countries omitted. Income classifications follow the World Bank’s 2025 classification. Standard errors clustered at the country level are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

Table F.3: Intangible Intensity and VC Volume Shares, Weights: log (1 + number of deals)

	(1) Total	(2)	(3)	(4) Total x Inc. Knowledge Knowledge x Inc. Organizational Organizational x Inc.	(5)	(6)
High intangible intensity (total)	0.430*** (0.027)	0.558* (0.029) -0.255*
× Upper-middle income		(0.056) -0.233***
High intangible intensity (knowledge)		(0.074)	0.083*** (0.028)	0.203*** (0.033)
× Low & lower-middle income				-0.199*** (0.055)
× Upper-middle income				-0.266*** (0.072)
High intangible intensity (organizational					0.451*** (0.025)	0.559*** (0.027)
× Low & lower-middle income						-0.219*** (0.052)
× Upper-middle income						-0.192*** (0.071)
Constant	0.170*** (0.013)	0.137*** (0.014)	0.343*** (0.015)	0.314*** (0.017)	0.159*** (0.012)	0.136*** (0.013)
Observations	483	483	483	483	483	483
R-squared	0.453	0.520	0.125	0.226	0.489	0.546
p-value: Low & LMI = 0		0.000		0.925		0.000
p-value: Upper-middle = 0		0.000		0.328		0.000

Notes: This table reports regressions of the share of VC volume across sectors on measures of sectoral intangible intensity over 2012–2023, using winsorized deal values. The dependent variable is the share of VC volume in each sector-country combination. Sectors are classified as meeting the relevant intangible-intensity definition, not meeting it, or lacking suficient information; regressions use the first two groups. Columns (1)–(2), (3)–(4), and (5)–(6) use the overall, knowledge, and orga nizational intangible intensity measures, respectively. Even-numbered columns interact intangible intensity with indicators for low- and lower-middle-income and upper-middle-income countries, with high-income countries omitted. Regressions are weighted by the log of one plus the country-level number of VC deals. Income classifications follow the World Bank’s 2025 classification. Standard errors clustered at the country level are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

Table F.4: Intangible Intensity and VC Volume Shares, Excluding China and India

	(1) Total	(2)	(3)	(4) Total x Inc. Knowledge Knowledge x Inc. Organizational Organizational x Inc.	(5)	(6)
High intangible intensity (total)	0.568*** (0.033)	0.617* (0.022) -0.269*
		(0.057) -0.247***
High intangible intensity (knowledge)		(0.072)	0.169***	0.239*** (0.031)
× Low & lower-middle income			(0.046)	-0.252*** (0.052)
× Upper-middle income				-0.405*** (0.071)
High intangible intensity (organizational)					0.544***	0.579*** (0.017)
× Low & lower-middle income					(0.025)	-0.205*** (0.054)
× Upper-middle income						-0.167** (0.069)
Constant	0.126*** (0.011)	0.111*** (0.007)	0.326*** (0.019)	0.300*** (0.013)	0.138*** (0.008)	0.130*** (0.005)
Observations	477	477	477	477	477	477
R-squared	0.846	0.879	0.538	0.683	0.856	0.880
p-value: Low & LMI = 0		0.000		0.767		0.000
p-value: Upper-middle = 0		0.000		0.011		0.000

Notes: This table reports regressions of the share of VC volume across sectors on measures of sectoral intangible intensity over 2012–2023, excluding China and India from the dataset. The dependent variable is the share of VC volume in each sector-country cell. Sectors are classified as meeting the relevant intangible-intensity definition, not meeting it, or lacking suficient information; regressions use the first two groups. Columns (1)–(2), (3)–(4), and (5)–(6) use the overall, knowledge, and organizational intangible intensity measures, respectively. Even-numbered columns interact intangible intensity with indicators for low- and lower-middle-income and upper-middle-income countries, with high-income countries omitted. Income classifications follow the World Bank’s 2025 classification. Standard errors clustered at the country level are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

Table F.5: Intangible Intensity and VC Volume Shares, Excluding Pre-seed and Seed Deals

	(1) Total	(2)	(3)	(4) Total x Inc. Knowledge Knowledge x Inc. Organizational Organizational x Inc.	(5)	(6)
High intangible intensity (total)	0.547*** (0.032)	0.608* (0.024) -0.264*
× Upper-middle income		(0.065) -0.154**
High intangible intensity (knowledge)		(0.067)	0.139***	0.225*** (0.035)
× Low & lower-middle income			(0.044)	-0.285*** (0.052)
× Upper-middle income				-0.239** (0.106)
High intangible intensity (organizational)					0.516*** (0.027)	0.567*** (0.020)
× Low & lower-middle income						-0.198*** (0.062)
× Upper-middle income						-0.136*** (0.045)
Constant	0.139*** (0.013)	0.114*** (0.008)	0.343*** (0.019)	0.305*** (0.014)	0.155*** (0.012)	0.134*** (0.006)
Observations	429	429	429	429	429	429
R-squared	0.825	0.848	0.507	0.596	0.831	0.850
p-value: Low & LMI = 0		0.000		0.125		0.000
p-value: Upper-middle = 0		0.000		0.890		0.000

Notes: This table reports regressions of the share of VC volume across sectors on measures of sectoral intangible intensity over 2012–2023. Pre-seed and seed deals are excluded; shares are computed over the remaining deals. The dependent variable is the share of VC volume in each sector-country cell. Sectors are classified as meeting the relevant intangible-intensity definition, not meeting it, or lacking suficient information; regressions use the first two groups. Columns (1)–(2), (3)– (4), and (5)–(6) use the overall, knowledge, and organizational intangible intensity measures, respectively. Even-numbered columns interact intangible intensity with indicators for low- and lower-middle-income and upper-middle-income countries, with high-income countries omitted. Income classifications follow the World Bank’s 2025 classification. Standard errors clustered at the country level are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

Table F.6: Knowledge and Organizational Keywords in Business Model

	(1)	(2)	(3)	(4)	(5)	(6)
		IT Sector			IT, Wholesale & Retail
	Knowledge	Organizational K minus O Knowledge			Organizational K minus O
Low & lower-middle income	-0.021	0.058***	-0.078***	-0.032**	0.094***	-0.125***
Upper-middle income	(0.016) -0.044***	(0.011) 0.045***	(0.026) -0.089***	(0.014) -0.047***	(0.014) 0.065***	(0.027) -0.112***
Constant	(0.011) 0.159*** (0.003)	(0.007) 0.102***	(0.017) 0.058***	(0.011) 0.150***	(0.012) 0.123***	(0.022) 0.027***
Observations	115,125	115,125	115,125	125,960	(0.004)	(0.006)
R-squared	0.001	0.003	0.003	0.001	125,960 0.006	125,960 0.006

Notes: The table reports firm-level regressions of keyword-based indicators on income-group dummies. Columns (1)–(3) restrict the sample to IT-sector firms, while columns (4)–(6) include firms in the IT, wholesale, and retail sectors. Keyword dictionaries include Chinese, Japanese, German, Hindi, Spanish, French, and Portuguese translations, with English terms retained. Columns (1) and (4) indicate whether a firm’s business description contains a knowledge-related keyword, including terms related to R&D, patents, algorithms, artificial intelligence, biotechnology, semiconductors, and frontier technology Columns (2) and (5) similarly capture organizational-related keywords, including terms related to distribution, logistics, marketplaces, supply chains, delivery networks, customer acquisition, e-commerce, mobile money, and digital payments Columns (3) and (6) report the diference between the knowledge and organizational indicators. Standard errors clustered at the country level are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

Table F.7: Levels and Variance Decomposition of ln(VC/pop) across Countries (2021-2023), Using Alternative Entry Definition

	Full sample	HICs	UMICs	Low & LMICs
Panel A: Entry, funding, and VC intensity by income group
Entry: first deals per million people (median)	1.06	8.69	0.82	0.21
Entry: first deals per million people (mean)	5.94	15.13	1.06	0.45
Funding Depth: USD M per entrant (median)	3.8	9.0	2.2	2.4
Funding Depth: USD M per entrant (mean)	6.3	9.4	6.1	3.3
VC/pop: USD per capita (median)	4.0	74.2	2.0	0.4
VC/pop: USD per capita (mean)	68.4	180.6	8.1	1.6
Countries	146	53	40	53
Panel B: Variance decomposition of ln(VC/pop)
Var( ln (VC/pop.))	10.2	2.9	4.6	6.0
Var( ln(First deals/pop.))
Var( ln(Funding Depth))	4.2 ( 41.5%) 2.3 ( 23.0%)	1.5 ( 50.2%) 0.5 ( 17.5%)	1.2 ( 26.7%) 2.4 ( 51.6%)	1.8 ( 29.9%) 2.7 ( 44.7%)
2*Cov( ln(First deals/pop.),ln(Funding Depth))	3.6 ( 35.5%)	0.9 ( 32.3%)	1.0 ( 21.7%)	1.5 ( 25.4%)
Countries	146	53	40	53

Notes: Panel A reports group medians and means for the level variables. Entry is defined as the number of firms receiving their first observed deal, of any type, per million inhabitants. Funding depth is computed using all deals other than each firm’s first and is defined as average total funding divided by the number of first deals (USD millions per entrant). VC/pop is total VC divided by population (USD per capita). Panel B uses the exact identity Var(ln VC/P) = Var(ln N /P) + Var(ln F D) + 2Cov(ln N₁/P, ln F D); values in parentheses report each component’s share of total variance (%). HICs = high-income countries; UMICs = upper-middle-income countries; LMICs = low- and lower-middle-income countries. HICs include the United States, UMICs include China, and LMICs include India. Income groups follow the World Bank’s 2025 classification.

Table F.8: Business Environment and the Entry–Depth Margin Balance (2021–2023)

VARIABLES	(1) Entry minus Depth coefficient	(2)
Minority investor protection (0–100)	0.000	0.011
× Low- & Middle-Income	(0.007)	(0.011) -0.003
WGI: Political stability (0-100)	0.032***	(0.015) 0.037***
× Low- & Middle-Income	(0.006)	(0.014) -0.019
Tax loss carryforward (years)	0.081***	(0.018) 0.050** (0.023)
× Low- & Middle-Income	(0.018)	0.033 (0.036)
Effective capital gains tax rate (pct)	0.014 (0.014)	0.006 (0.021)
× Low- & Middle-Income		0.001 (0.029)
Corporate income tax rate (pct)	-0.069*** (0.018)	-0.044** (0.019)
× Low- & Middle-Income		-0.012 (0.037)
Log equity issuance / GDP	-0.038** (0.015)	0.001 (0.022) -0.075**
× Low- & Middle-Income Low- & Middle-Income		(0.034) 0.494
Constant	-2.051**	(1.888) -2.989** (1.360)
	(0.830)
Observations	121	121
R-squared	0.406	0.455

Notes: This table reports OLS regressions relating country characteristics to the entry–depth margin, defined as the diference between the entry and funding-depth components of VC activity. Entry is defined as in the main text. Because ln(V C/pop) = ln(entry/pop) + ln(funding depth), positive coeficients indicate a stronger association with entry, while negative coeficients indicate a stronger association with funding depth. Entry and funding depth are winsorized at the top and bottom 1% before forming the diference. Column (2) interacts each characteristic with an indicator for low- and middle-income countries; main efects therefore correspond to high-income countries. Minority investor protection and WGI political stability are min–max normalized to a 0–100 scale. Estimates reflect cross-country associations and are not causal. Robust standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

Table F.9: Business Environment and VC Ecosystem Margins (2021–2023), Using Alternative Entry Definition

	(1)	(2)	(3)
VARIABLES			ln(VC/pop) ln(First deals/pop) ln(Funding Depth)
Minority investor protection (0–100)	0.033**	0.020***	0.013*
WGI: Political stability (0–100)	(0.013) 0.047***	(0.006) 0.040***	(0.008) 0.008
Tax loss carryforward (years)	(0.009) 0.093***	(0.005) 0.080***	(0.005) 0.013
Effective capital gains tax rate (pct)	(0.024) 0.001	(0.014) 0.004	(0.014) -0.003
Corporate income tax rate (pct)	(0.024) -0.025	(0.013) -0.043***	(0.014) 0.018
Log equity issuance / GDP	(0.028) 0.090***	(0.016) 0.025*	(0.017) 0.066***
Constant	(0.025) -3.098**	(0.013) -3.037***	(0.015) -0.061
	(1.356)	(0.705)	(0.809)
Observations	121	121	121

Notes: Entry is defined as the number of firms receiving their first observed deal, of any type, per million inhabitants. Minority investor protection and WGI political stability are min–max normalized to a 0–100 scale. Column (1) is constructed as the sum of the winsorized entry and funding-depth components, so the coeficients in columns (2) and (3) sum exactly to those in column (1). Funding depth is computed over all subsequent deals. Robust standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

Table F.10: Business Environment and VC Ecosystem Margins by Sector (2021–2023), Using Alternative Entry Definition

	(1)	(2)	(3)	(4)	(5)	(6)
	Knowledge-intensive		Organizational-intensive		Traditional
	ln(entry/pop)	ln(FD)	ln(entry/pop)	ln(FD)	ln(entry/pop)	ln(FD)

Minority investor protection (0–100)	0.012*	-0.002	0.009	0.015	0.014**	0.001
WGI: Political stability (0–100)	(0.007)	(0.008)	(0.007)	(0.012)	(0.006)	(0.016)
WGI: Political stability (0–100)	0.045***	0.009	0.044***	0.000	0.035***	0.005
Tax loss carryforward (years)	(0.006)	(0.007)	(0.006)	(0.008)	(0.005)	(0.008)
Tax loss carryforward (years)	0.096***	0.030*	0.091***	0.026	0.087***	0.031
Effective capital gains tax rate (pct)	(0.016)	(0.017)	(0.015)	(0.022)	(0.012)	(0.027)
Effective capital gains tax rate (pct)	-0.003	-0.015	-0.006	-0.007	-0.013	-0.034
Corporate income tax rate (pct)	(0.012)	(0.014)	(0.010)	(0.016)	(0.008)	(0.022)
Corporate income tax rate (pct)	-0.055***	0.025	-0.040**	0.025	-0.023*	0.006
Log equity issuance / GDP	(0.016)	(0.020)	(0.016)	(0.025)	(0.013)	(0.032)
Log equity issuance / GDP	0.024*	0.068***	0.006	0.095***	0.000	0.088***
Constant	(0.014)	(0.021)	(0.013)	(0.024)	(0.013)	(0.029)
Constant	-3.133***	0.684	-4.060***	-0.119	-4.972***	1.746
	(0.689)	(0.941)	(0.756)	(1.398)	(0.677)	(1.312)

Observations R-squared	100	100	100	100	100	100
	0.742	0.342	0.698	0.432	0.734	0.245

Notes: OLS regressions with robust standard errors. Each column restricts the dependent variable to deals in the indicated sector (K = knowledge-intensive; O = organizationally-intensive; T = traditional). Entry is defined as a firm’s first observed deal, of any type; funding depth is computed over all subsequent deals. The sample is restricted to countries with non-missing values on all six outcomes and all six regressors in 2021–2023. Outcomes are winsorized at the top and bottom 1%. Minority investor protection and WGI political stability are min–max normalized to a 0–100 scale. Cross-sector equality of coeficients is tested via one-sided Wald tests on stacked equations using a robust joint covariance matrix; the results are reported in the accompanying log. Robust standard errors are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

What Does Venture Capital Actually Finance? Scale and Structure around the World

常见问题