## wpiea2024152-print-pdf

## Source details

**Canonical URL:** [wpiea2024152-print-pdf](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024152-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024152-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024152-print-pdf.pdf.json)

---

### Introduction, research question, dataset, and methods
- Research focus: empirically examine the extent to which both inward and outward FDI influence global knowledge diffusion using patent citations as a direct measure of knowledge flows.
- Dataset construction and coverage:
  - Combines project-level greenfield investment data and cross-border mergers and acquisitions (brownfield FDI) with worldwide patent citation data from PATSTAT through name-matching.
  - Database includes transactions from 12,656 firms and bi-directional patent citations involving 60 countries over the period 2003-2022.
  - The 60 countries are the top 60 countries by the number of granted patents in PATSTAT, collectively representing 99.9% of the granted patents in this dataset.
- Methodology:
  - Applies the local projection difference-in-differences (LP-DiD) methodology developed by Dube et al. (2023) to address heterogeneous dynamic treatment effects in staggered DiD settings.
  - Uses extensive fixed effects to control for time-varying factors at host-country(-industry) and firm levels, including variation in innovation rates, citation propensities, and patent activities.
- Main empirical outcomes (summary):
  - After a firm’s initial entry into a country, citations by the host country towards the investing firm increase by around 7.8% to 10.6%, depending on the investment type.
  - Investing firms increase their citations towards the host country by around 4.5% to 13%.
  - Results robust to a range of alternative specifications and definitions.

### Data sources, citation data, and linking to FDI
- Citation data (PATSTAT Spring 2022):
  - PATSTAT contains bibliographic details on patents granted by or applied to 90 patent-issuing authorities.
  - Analysis considers only citations between firms (companies as opposed to patents owned by individual inventors).
  - Triadic patents used in some analyses as a high-quality indication.
- Company identification and patent counting:
  - EPO provides name disambiguation and EPO IDs; applicants labeled as “company” are retained.
  - Citations measured at the patent-family level; only applicant-made citations included; examiner-added citations excluded.
  - Fractional counting: for a patent with N IPC codes and M companies assign 1/(NM) patents to each firm-IPC pair; firm-ISIC counts derived via probabilistic IPC→ISIC crosswalk.
- Linking patents to FDI:
  - Name standardization and string-similarity matching (cosine TF-IDF tri-gram and Dutch central bank algorithm); threshold similarity score 0.7 used; manual checks for mismatches.
  - “World sample”: matches from this procedure.
  - “US sample”: builds on PATSTAT–U.S. Compustat matches (Arora et al. (2021b)) with online search similarity refinement (requires at least one common webpage among the first ten Bing results).
- Definitions of citation measures:
  - Firm citations from country c in year t: sum of fractional citations received by all patents belonging to a firm f in that year.
  - Reverse citations: sum of fractional citations made by all (granted) patents applied in year t by the investing firm to patents in country c.
  - Primary outcome variable: stock of citations (cumulative citations), starting 1995.
- Sample scope and counts:
  - Worldwide sample coverage:
    - 12,656 firms globally.
    - 12,696 brownfield FDI projects.
    - 4,632 acquiring firms.
    - 87,415 greenfield FDI projects by 10,096 investing firms.
  - US merged sample coverage:
    - 1,872 firms.
    - 1,143 firms engaged in a total of 2,850 M&A deals overseas.
    - 1,438 firms invested in a total of 16,433 greenfield investment projects.
  - Analysis limited to top 60 destinations by granted patents (2003-2022), representing over 99.9% of all patents.

### Stylized facts and empirical strategy
- Time-series stylized facts:
  - Brownfield FDI dominated by ADV⇒ADV flows; brownfield increased until the global financial crisis then plummeted and slowly recovered.
  - Greenfield ADV⇒DEV outpaced ADV⇒ADV until the crisis; post-crisis ADV⇒DEV stagnated while ADV⇒ADV grew until the pandemic.
  - Cross-country patent citation evolution predominantly driven by ADV⇒ADV; developing countries began significantly citing advanced-country patents only about a decade ago.
- Residualized firm-destination correlation:
  - PPML residualization a la Silva and Tenreyro (2006) with firm-year, destination country-year, and country-pair fixed effects.
  - Binned-scatter evidence: investor firm patents are cited more frequently from countries where the firm invested more; firms cite destination-country patents more when they invest more—correlation persists after fixed effects.
- LP-DiD event-study framework and identification:
  - Baseline event-study: y_{ic,t+h} − y_{ic,t−1} = β_h D_{ic,t} + Σ γ_hk y_{ic,t−k} + η′ x_{ic,t−1} + δ^h_{it} + δ^h_{ct} + ε^h_{ic,t}.
  - Treatment D_{ic,t} indicates first year firm i invests in country c during 2003-2022.
  - Stabilization lag L = 5 (baseline) to exclude contamination from prior/subsequent treatments; L = 5 chosen because 95 percent of subsequent investments occur within 5 years.
  - Controls include three lags (p = 3) of the asinh cumulative citations and trend y_{t−4} − y_{t−6}; robustness to alternative lag structures reported.
  - Dependent variable: change in asinh(cumulative citations) from 1995 onward; results robust to log(1 + x) and triadic-patent specifications.
  - Parallel-trend tests: pre-treatment β_h coefficients reported; alternative specifications (firm-country-industry, dropping citations from acquired targets) used to address pre-trends and mechanical issues.

### Main quantified effects (Section 4.1: baseline world sample and US sample)
- Baseline world-sample five-year effects (asinh change translated to percentage changes using sinh):
  - Brownfield FDI: five years after brownfield investment destination countries increase asinh cumulative citations to investing firms by 0.063.
    - Pre-treatment (t−1) average asinh cumulative citations: 1.158 (corresponds to 1.43 untransformed cumulative citations).
    - Implied increase: around 7.8% in citations to investing firms from the FDI destination country.
    - 95% confidence interval: 5.7−10%.
  - Greenfield FDI: implied percentage change in citations from greenfield FDI: 10.6%.
    - 95% confidence interval: 9−12.2%.
- Investors’ citations to host countries (baseline):
  - Reported increases: 10.8% and 13.4% in different figures.
  - Greenfield specification displays a small pre-trend (trend coefficient 0.0067 asinh cumulative citations per year for coefficients 4 and 5 years before entry); extrapolating subtracts this trend and implies an 11.8% increase five years after treatment.
- US sample (Arora et al. (2021b) matched sample) estimates:
  - Citations received by investing firms: increases of 8.4% and 9.9% (Figures IVa and IVb).
  - Investors’ citations to host countries: increases of 12.7% and 11.1% (Figures IVc and IVd).
  - Slight pre-trend for greenfield FDI in US sample; no parallel-trend violation for brownfield FDI.
- Patent reassignment sensitivity (US sample):
  - Restricting to patents originally assigned to US firms does not significantly affect country→investor citation estimates.
  - The same restriction yields significantly lower measured investor→country citation effects, indicating part of the investor-side effect is mechanical reassignment of destination-country patents to investors in Arora et al. (2021b) dataset.
- Overall baseline summary:
  - Midpoints range from 7.8% to 12.7% increases in citations flowing between investors and host countries.
  - Greenfield shows larger implied percentage changes but detectable pre-trends; firm-country-industry specifications and robustness checks yield qualitatively similar but sometimes smaller effects.

### Scope of diffusion across untargeted firms and country-industry (Section 4.2)
- Two refined specifications:
  - (i) exclude citations originating from firms acquired in brownfield FDI to address self-citations.
  - (ii) analyze at country-industry (ISIC 2-digit) level including firm-industry-time and country-industry-time fixed effects and technological linkages.
- Dropping citations from acquired targets (brownfield):
  - Host country’s citations to brownfield investors:
    - World sample: 8.9%.
    - US sample: 7.3%.
    - These bounds cover baseline 7.8% and 8.4%.
  - Investors’ citations to host countries:
    - World sample: 10.8% (same as baseline).
    - US sample: 4.5% (smaller than baseline; suggests about two thirds of the effect in Figure IVc is due to citations made by investors to acquired firms in destination countries).
- Firm-industry specification and knowledge input coefficients:
  - Knowledge input coefficient a_{c l←k} defined using 1990-2000 citations, ranges 0–1, sums to 1 across k.
  - Treatment intensity T_{icl,t} ≡ Σ_k D_{ick,t} × a_{c l←k}, scaling treatment by technological relatedness.
- Industry-level greenfield results (top 30 destinations; US sample; ≈276 million observations):
  - Average increase in citations made by host industries to greenfield investors: 9%.
  - Investors’ citations to target industries: 10.8%.
  - Input-coefficient interaction:
    - Theoretical extreme a_{c l←k} = 1 implies a 32.7% increase in citations by related industries to greenfield investors (not empirically observed).
    - Same-industry share a_{l←l} = 22.4% implies a 7.2% increase for directly targeted sectors.
    - Average T_{icl,t} among industries with T_{icl,t} > 0 is 5.7%, implying a 1.8% increase in citations made to investors by sectors connected to FDI targets.
    - Non-targeted related sectors can experience larger percent increases (e.g., 28.5% for non-targeted industries at a_{l←k} = 1 vs 22.7% for targeted sectors in interaction specification).
  - Coverage:
    - Directly targeted greenfield industry-country pairs: 3,487.
    - Treated industry-country pairs under knowledge-input specification: 126,598 (3,487 equals 2.75% of 126,598).
- Industry-level brownfield results (limitations):
  - Primary-industry assignment for brownfield limits precision; target and investor primary industries mismatch in 75% of cases.
  - Using acquiring firms’ primary industry: positive but not significant effects for citations from invested industries to investor firms; imprecise zero effects for investor→country citations.
  - Using knowledge input coefficients recovers spillovers similar to greenfield case.
  - Assigning target-firm primary industry: positive effects from target country-industry→investor firms but no effect investor→target country-industry.
- Interpretation: knowledge spillovers extend beyond directly targeted firms and industries, frequently benefiting technologically related sectors; measurement issues larger for brownfield at industry granularity.

### Heterogeneity by host-country absorptive capacity (Section 4.3)
- Absorptive-capacity measures:
  - Destination patent-stock dummy: I_{ic,t} = 1 if destination c belongs to top 10% of destinations by patent quantity (1990-2000 reference); top five: U.S., China, Japan, Germany, Korea.
  - Technological similarity: cosine similarity between investor and country patent-stock vectors across ISIC 2-digit industries over previous ten years; multiplied by 10 so I_{ic,t} ranges 0–10; sample mean 1.26.
- Heterogeneity by destination patent stock (top 10% vs bottom 90%):
  - Countries’ citations to investors (five-year effects):
    - Brownfield: bottom 90% increase 5.1%; top 10% increase 12.3%.
    - Greenfield: bottom 90% increase 2.73%; top 10% increase 21.3%.
  - Investors’ citations to host countries:
    - Brownfield: bottom 90% increase 6.8%; top 10% increase 17.6%.
    - Greenfield: bottom 90% increase 4.9%; top 10% increase 24.5%.
  - Interpretation: being among top patent-producing destinations amplifies citation-flow effects by roughly twofold or more.
- Heterogeneity by technological similarity:
  - Higher cosine similarity yields larger citation flows in both directions.
  - Countries can benefit even at very low technological similarity; investors’ citations to hosts require some similarity (greenfield: zero similarity implies no investor→host citations).
  - Quantitative examples:
    - Greenfield host-country citations: 6.7% at 25th percentile and 9.2% at 75th percentile (reported).
    - Alternative textual report: citations increase by 4% at 25th percentile and 11.1% at 75th percentile in a different specification.
    - For brownfield, moving from 25th to 75th percentile of similarity increases citations by about 32%; same move more than doubles greenfield effect.
  - Note: percentage effects may be lower at higher similarity because pre-treatment citation levels are higher there; absolute changes are larger at higher similarity.
- Other heterogeneity and robustness notes:
  - No conclusive evidence on FDI effects on patent creation due to identification challenges; brownfield effects on granted vs triadic patents diverge (negative vs positive).
  - Robustness checks where baseline results remain robust include:
    - (1) expanding destination set to all PATSTAT countries;
    - (2) limiting forward citations to five years after application;
    - (3) increasing stabilization lags L to 12;
    - (4) using log(1 + x) instead of asinh;
    - (5) using triadic patents only;
    - (6) treating each investment as separate event;
    - (7) grouping greenfield and brownfield into single FDI variable.
  - Appendix A robustness midpoints and ranges (baseline reported with 95% C.I. in parentheses):
    - (a) Citations to brownfield investors increase on average by 7.8% (5.7−10%).
    - (b) Citations to greenfield investors increase on average by 10.6% (9−12.2%).
    - (c) Citations to countries made by brownfield investors increase by 10.8% (8.4−13.3%).
    - (d) Citations to countries made by greenfield investors increase by 13.4% (11.6−15.2%).
  - Robustness alternative outcomes (selected):
    - Using triadic patents only: (a) 6.6%; (b) 9.5%; (c) 10%; (d) 9.5%.
    - Using log(1 + x): (a) 9.4%; (b) 12.7%; (c) 13.1%; (d) 16.2%.
    - Using only patents originally assigned to US firms (US sample): (a) 5.7%; (b) 9.1%; (c) 3.5%; (d) 5.4%.

### Key implications and conclusions
- Average estimated citation changes after first investor entry:
  - Destinations’ citations to investors: increase by 7.8%–10.6% on average, depending on investment type.
  - Investors’ citations to FDI hosts: increase by 4.5%–13%.
- Spillovers reach beyond targeted firms and industries, affecting technologically related sectors and sometimes yielding larger percent gains in related non-targeted sectors.
- Host-country absorptive capacity matters strongly:
  - Hosts with larger pre-existing patent stocks and higher technological similarity to investing firms reap substantially larger citation flows.
- Policy and research implications highlighted by the authors:
  - Sizable knowledge flows induced by FDI imply potential productivity and growth losses if investment fragmentation occurs amid geopolitical tensions.
  - Destinations’ capacity to reap FDI benefits is critical; further investigation needed into channels of knowledge transmission, heterogeneity by investment type, classification of FDI motives via text analysis, diffusion of non-patent organizational knowledge, and whether acquired firms/industries increase overall patenting.

*Source: wpiea2024152-print-pdf*

### References .............................................................................................................

### References

### Introduction: research question, dataset, and methods
- Research focus: empirically examine the extent to which both inward and outward FDI influence global knowledge diffusion using patent citations as a direct measure of knowledge flows.
- Dataset construction:
  - Combines project-level greenfield investment data and cross-border mergers and acquisitions (brownfield FDI) with worldwide patent citation data from PATSTAT through name-matching.
  - Database includes transactions from 12,656 firms and bi-directional patent citations involving 60 countries over the period 2003-2022.
  - The 60 countries are the top 60 countries by the number of granted patents in PATSTAT, collectively representing 99.9% of the granted patents in this dataset.
- Methodology:
  - Applies the local projection difference-in-differences (LP-DiD) methodology developed by Dube et al. (2023) to address heterogeneous dynamic treatment effects in staggered DiD settings.
  - Uses extensive fixed effects to control for time-varying factors at host-country(-industry) and firm levels, including variation in innovation rates, citation propensities, and patent activities.
- Main empirical outcomes:
  - After a firm’s initial entry into a country, citations by the host country towards the investing firm increase by around 7.8% to 10.6%, depending on the investment type.
  - Investing firms increase their citations towards the host country by around 4.5% to 13%.
  - Results are robust to a range of alternative specifications and definitions.

### Contributions to the literature
- Methodological:
  - Introduces LP-DiD to FDI–knowledge spillover analysis to better identify causal impacts in staggered treatment settings.
- Novelty in FDI type comparison:
  - Leverages both greenfield and brownfield investment data to compare their differential impacts on knowledge spillovers within a unified empirical framework.
- Granularity and scope:
  - Global firm-level dataset enables exploration of heterogeneity across countries and industries, absorptive capacity along the technological dimension, cross-industry spillovers, and bi-directional knowledge flows (host-to-investor and investor-to-host).

### Key findings and heterogeneity patterns
- Bi-directional knowledge exchange:
  - Significant and roughly equal increases in citations between host countries and investment firms after initial entry.
- Greenfield vs brownfield:
  - Patent citations tend to increase more following greenfield compared to brownfield investment, though the difference is modest.
- Cross-industry and technologically related effects:
  - Spillovers extend beyond directly targeted firms and industries, affecting technologically related sectors; in some cases, related sectors benefit more—percentage-wise—than directly targeted sectors.
- Role of absorptive capacity and technological similarity:
  - Interactions with host countries that have larger pre-existing patent stocks result in spillovers that are two to ten times greater.
  - Technological similarity between investing firms and host countries facilitates knowledge spillovers; when technological similarity is low, spillovers appear insignificant.

### Related literature themes and contrasts
- Historical context:
  - MNCs account for around 90% of total exports and imports (Bernard et al., 2009) and are central to FDI flows.
  - Cross-border patent citations have surged in recent decades (IMF, 2018; LaBelle et al., 2023).
- Prior studies:
  - Country- or region-specific findings showing FDI-related citation increases (e.g., Branstetter (2006) for Japanese firms in the U.S.; Globerman et al. (2000) for Swedish outward FDI).
  - Recent evidence: Akcigit et al. (2024) find foreign corporations investing in U.S. startups increase their own citations to those startups post-investment.
- Absorptive capacity literature:
  - Borensztein et al. (1998): countries with higher human capital benefit more from inward FDI.
  - Alfaro et al. (2004): more developed financial systems enable better exploitation of inward FDI.
  - This paper specifically explores absorptive capacity along the technological dimension.
- Greenfield vs brownfield motivations and implications:
  - Theoretical and empirical work on differing motivations: greenfield may preserve firm-specific advantages (Chen and Zeng, 2004); M&A may target innovation or serve non-innovation purposes such as reducing competition or tax strategies (Bena and Li, 2014; Phillips and Zhdanov, 2013; Cunningham et al., 2021; Belz et al., 2013).
  - Empirical evidence on how these differing motivations affect knowledge diffusion remains limited; this paper addresses that gap.

### Data sources and descriptions
- Greenfield FDI (fDi Markets):
  - Source: fDi Markets (fDi Intelligence, Financial Times Group).
  - Coverage: investment-level information for over 300,000 FDI deals between January 2003 and December 2022.
  - Data elements: parent company name, source and destination countries, industrial sector, activity type (e.g., business services, sales, R&D), investment category (new investment or expansion), value of investments, estimated number of jobs created.
  - Data collection: primarily from publicly available sources (media sources, industry organizations, investment promotion agencies news wires).
  - Reliability: confirmed by aggregating FDI values at destination country-year level and comparing with official gross FDI inflows (e.g., Toews and V ́ezina, 2022; Aiyar et al., 2023).
- Brownfield FDI (Refinitiv Eikon):
  - Source: Refinitiv Eikon (formerly SDC Platinum by Thomson Reuters).
  - Coverage: detailed information on cross-border M&A transactions representing the acquisition of at least a 5% stake or of a 3% stake with a deal value of at least USD$1 million, covering more than 1.45 million deals in the world since the 1970s.
  - Data elements: acquirer and target firm names and locations, sector associated with the target firm, purchase value (in USD).
  - Usage: primary source for cross-border M&A patterns reported in World Investment Report by UNCTAD; extensively used in prior empirical research (e.g., Erel et al. (2022); Bergant et al. (2023)).

*International Monetary Fund — IMF WORKING PAPERS: Knowledge Diffusion Through FDI: Worldwide Firm-Level Evidence*

### 2.2    Citation Data

### 2.2    Citation Data

### Data source and coverage
- Citation data obtained from the Spring 2022 release of PATSTAT, maintained by the European Patent Office (EPO).
- PATSTAT contains bibliographic details on patents granted by or applied to 90 patent-issuing authorities.
- Analysis considers only citations between firms (companies as opposed to patents owned by individual inventors).
- Country associated with each patent determined from the address reported by the firm to the patent authorities at the time of application.
- Triadic patents (patents registered in the US Patent Office (USPTO), Japanese Patent Office, and EPO) are used in some analyses as a high-quality indication.

### Company identification and country attribution
- The EPO performs name disambiguation to assign unique EPO IDs to applicants by consolidating different versions of company names.
- EPO labels applicants to indicate if they are a “company”; analysis focuses on applicants classified as companies, excluding inventors.
- When an EPO ID has multiple self-reported addresses in different countries, the country most frequently associated with that EPO ID is attributed.
- If no self-reported addresses are associated with an EPO ID, the country corresponding to the patent-granting authority in which the EPO ID occurs most frequently is attributed.
- Note on multiple country designations: these may reflect local subsidiaries reporting local addresses rather than headquarters.

### Patent classification and high-quality patent indicators
- International Patent Classification (IPC) is mapped into International Standard Industrial Classification (ISIC) via a probabilistic crosswalk (Lybbert and Zolas, 2014).
- Triadic patents are used in some analyses as higher-quality indications (de Rassenfosse et al., 2014).

### Citation counting methodology
- Citations measured at the level of a “patent family” to avoid double counting.
- Only citations made by patent applicants are included; examiner-added citations are excluded to better reflect knowledge spillovers from inventors at time of invention.
- Fractional counting of patents and citations:
  - For a patent applicable to N IPC codes and belonging to M companies, assign 1/(NM) patents to each firm-IPC pair.
  - Firm-ISIC counts obtained by applying the probabilistic IPC→ISIC crosswalk to firm-IPC counts.
  - Aggregation: sum relevant fractional counts (e.g., a country-level count of a patent registered in multiple countries equals the sum of each country’s assignees over the total number of assignees).

### 2.3    Linking Citations and FDI Information

### Name matching and sample construction
- Challenge: align names across FDI datasets and PATSTAT given over 107,000 and 541,000 unique names in each dataset, respectively.
- Standardize names following procedures described by Arora et al. (2021b).
- Restrict PATSTAT to firms with five or more patents, reducing unique name count to 100,000.
- Apply string similarity metrics:
  - Cosine similarity with TF-IDF weighting at the tri-gram character level.
  - A matching algorithm developed by the Dutch central bank.
- Among firm-name pairs with a similarity score over the threshold of 0.7 (out of 1) from either metric, mismatches are filtered by manual checking of firm names.
- Matches from this procedure are referred to as “World sample”.

### U.S. sample construction and improvements
- Concerns over global string-matching accuracy addressed by creating a U.S. investing-firms dataset with well-established matches from previous studies.
- Builds on PATSTAT–U.S. Compustat firm name matches developed by Arora et al. (2021b), which:
  - Assign patents to their ultimate owners and accommodate changes in ownership, covering patents from 1980 to 2015.
  - Reassign patents over time when firms undergo changes in ownership structure.
- U.S. matching additionally uses online search similarity (following Autor et al., 2020) and requires at least one common webpage among the first ten Bing search results to refine matches. This yields higher accuracy for both patent and FDI information.
- This subset is referred to as “US sample.”

### Definitions of citation measures and analytic focus
- Firm citations from country c in year t: sum of fractional citations received by all patents belonging to a firm f in that year.
- Reverse citations: sum of fractional citations made by all (granted) patents applied in year t by the investing firm to patents in country c.
- These two measures proxy knowledge flows from investing firms toward destination countries and vice versa.
- Primary outcome variable: stock of citations (cumulative citations), chosen over flow citations due to sporadic nature of flows and clearer interpretation of cumulative knowledge spillovers over time after FDI.

### Sample restrictions for computational tractability
- Analysis limited to the top 60 destinations by number of granted patents from 2003-2022, representing over 99.9% of all patents.
- Rationale: manage dataset size, especially for firm-country-sector-year analysis that otherwise would exceed 180 million observations.
- Overall worldwide sample coverage:
  - 12,656 firms globally.
  - 12,696 brownfield FDI projects.
  - 4,632 acquiring firms.
  - 87,415 greenfield FDI projects by 10,096 investing firms.
- Merged U.S. sample coverage:
  - 1,872 firms.
  - Among these, 1,143 firms engaged in a total of 2,850 M&A deals overseas.
  - 1,438 firms invested in a total of 16,433 greenfield investment projects.

*Source: wpiea2024152-print-pdf - 2.2    Citation Data*

### 2.4    Stylized Facts

### 2.4    Stylized Facts

### Time-series evolution of FDI and patent citations
- Countries are grouped into advanced and developing; four flows traced: ADV⇒ADV, ADV⇒DEV, DEV⇒ADV, DEV⇒DEV. Each series normalized to ADV⇒ADV flow value in 2003 (index baseline).
- Brownfield FDI (panel Ia):
  - Dominated by ADV⇒ADV flows, followed by ADV⇒DEV flows.
  - Worldwide brownfield FDI increased rapidly up until the global financial crisis when it plummeted.
  - Since the crisis, brownfield FDI has been slowly recovering to reach the pre-crisis peak only recently.
  - Investment flows originating from developing countries account for only a minor share of total brownfield FDI.
- Greenfield FDI (panel Ib):
  - Greenfield FDI from advanced to developing countries outpaced greenfield between advanced countries until around the global financial crisis.
  - After the crisis, greenfield ADV⇒DEV stagnated while greenfield ADV⇒ADV continued growing until the pandemic.
  - Investment flows originating from developing countries account for only a minor share of total greenfield FDI.
- Patent citations (panel Ic):
  - Cross-country patent citation evolution over the past two decades is predominantly driven by citations between advanced countries (ADV⇒ADV).
  - Developing countries began to cite patents belonging to advanced countries increasingly at a notable level only about a decade ago.

### Relationship between FDI and patent citations (firm-destination level)
- Residualization strategy:
  - FDI and patent citation measures are residualized using Poisson pseudo-maximum likelihood (PPML) estimator `a la Silva and Tenreyro (2006)` with fixed effects:
    - FE_it : firm-year fixed effects.
    - FE_ct : destination country-year fixed effects.
    - FE_oc : country-pair fixed effects.
  - Equation form: Y_ioct = exp[FE_it + FE_ct + FE_oc] × ε_ioct, where Y_ioct is either firm i’s FDI transactions to country c in year t (brownfield or greenfield) or patent citation counts from country c to firm i in country o in year t.
- Binned-scatter evidence (Figure II):
  - Top row: patent citations measured as total citations made by new patents from a destination country to patents belonging to an investor firm.
  - Bottom row: patent citations measured as total citations made by new patents by an investor firm to patents belonging to a destination country.
  - Left column: brownfield FDI; right column: greenfield FDI.
  - Both axes are residualized against source-destination country pair, destination country-year, and firm-year fixed effects by PPML.
- Main empirical finding:
  - For both brownfield and greenfield FDI:
    - An investor firm’s patent tends to be cited more frequently from a country where the firm invested more.
    - The firm is more likely to cite patents belonging to a country with greater investment.
  - This positive correlation persists after purging firm-, country-, and country-pair-specific confounding effects, motivating causal identification of the direction between FDI and citations.

### Empirical strategy (LP-DiD framework and identification)
- Methodology:
  - Use local projection difference-in-differences (LP-DiD) framework for event studies (Dube et al., 2023).
  - Advantages cited:
    - Accommodates control for pre-treatment variables (unlike standard DiD).
    - Avoids bias from heterogeneous treatment effects across treatment groups through sample selection.
  - LP-DiD restricts sample so treatment and control groups are unaffected by delayed treatment effects from previous events.
- Baseline event-study specification:
  - Outcome: y_ic,t+h − y_ic,t−1 = β_h D_ic,t + sum_{k=1}^p γ_hk y_ic,t−k + η' x_ic,t−1 + δ^h_it + δ^h_ct + ε^h_ic,t.
  - D_ic,t is a dummy indicating firm i invested in country c in year t for the first time over 2003-2022.
  - δ^h_it and δ^h_ct are firm-year and destination-year fixed effects; x_ic,t−1 are other controls.
  - β_h identifies the cumulative impact of FDI on citations measured h years after the event.
- Sample restriction to avoid contamination (Equation (3)):
  - Include observations that are either:
    - investment episodes: D_ic,t = 1, D_ic,t−j = 0 for 1 ≤ j ≤ L, or
    - clean controls: D_ic,t−j = 0 for −h ≤ j ≤ L.
  - Purpose: ensure neither treatment nor control groups have been treated during the L periods preceding the FDI event and to exclude observations treated during t−L to t+h to avoid delayed effect contamination.
- Parallel trends and mitigation:
  - Causal interpretation requires parallel trends conditional on controls.
  - Controls and unit of analysis chosen to mitigate violations:
    - Firm-time fixed effects control for inherently more productive/innovative firms.
    - Host country-time fixed effects account for targeting of more innovative countries.
    - High-level aggregation of FDI destination reduces endogeneity from investor selective choice.
  - Additional steps discussed to lessen endogeneity: excluding citations from subsidiaries, focusing on industries not directly affected by FDI.
  - Pre-treatment β_h coefficients reported to test for parallel trends.
- Treatment definition and robustness:
  - Baseline: initial entry of a firm into a destination during 2003-2022 as treatment (D_ic,t indicates first entry).
  - Rationale: follows Dube et al. (2023) for non-absorbing treatments; assumes subsequent FDIs are potentially connected to the initial one.
  - Empirical note: 95 percent of subsequent investments take place within 5 years for any given firm-country pair with FDI entry.
  - Set stabilization lags L = 5, which leads to exclusion of these 95 percent subsequent investments from control and treatment groups.
  - Robustness checks:
    - Alternative definition where D_ic,t is any FDI (first and subsequent) yields statistically indistinguishable results.
    - Separate regressions estimated for brownfield and greenfield FDI; combined-specification also considered with no significant differences reported.
- Dependent variable:
  - Change in the inverse hyperbolic sine (asinh) of cumulative citations from 1995 onwards used as main outcome.
  - Rationale for 1995 start: earliest year for aligning M&A and patent data.
  - asinh chosen to accommodate zeros and negative values; coefficients approximated and converted back to percentage changes for interpretation.
  - Robustness:
    - Results robust to log(1 + x) transformation.
    - Results robust when focusing on triadic patents (registered with USPTO, JPO and EPO); effects are qualitatively unchanged and significantly higher when using the log transformation.
    - Alternative specification restricting citations to within five years after initial patent application (excluding final five years) yields broadly consistent results.
- Choice of stabilization lags L:
  - Baseline set to L = 5.
  - Considerations:
    - Trade-off: longer L lowers bias but increases variance due to fewer events.
    - Setting L = 5 allows inclusion of "clean" events from 2003 + L = 2008 onwards.
    - With L = 5 and the sample span, effects can be estimated for up to 14 periods after treatment.
    - PATSTAT citation data reliable only up to 2020 due to reporting lags, implying a maximum feasible L = 12.
    - US sample limitations (Arora et al. (2021b) database ends in 2015) constrain maximum consistent L to 6.
    - Previous version used L = 6 with largely unchanged results but fewer events for longer horizons.
    - Robustness checks also verify results with L = 12.
- Choice of controls and lag structure:
  - Baseline includes three lags of the asinh cumulative citations (p = 3) and the trend in this variable between 4 and 6 periods before the period considered, y_{t−4} − y_{t−6}.
  - Justification: observable differences in levels or trends detected in some settings; this specification broadly removes significant differences in pre-trends in most cases.
  - Previous experiments: p = 2, dropping y_{t−4} − y_{t−6} and replacing it with y_{t−3} − y_{t−5}, producing qualitatively similar results.

*Source: wpiea2024152-print-pdf - 2.4    Stylized Facts*

### 4.1    Main Result:  Citation Flows Between Host Countries and Investing Firms

### 4.1    Main Result:  Citation Flows Between Host Countries and Investing Firms

### Baseline specification and estimation approach
- Dependent variable: change in the inverse hyperbolic sine (asinh) of cumulative citations received by (or made by) firm i from (to) country c since 1995.
- Controls: firm-year and country-year fixed effects, three lags of the dependent variable in levels, and its trend based on changes 4 to 6 years prior to the event.
- Stabilization period L set to 5 years.
- Estimated coefficients β_h are displayed for the year of first entry of firm i in country c during the sample period 2003-2022.

### Main quantified effects (baseline, world sample)
- Brownfield FDI: five years after a brownfield investment, destination countries increase asinh cumulative citations to investing firms by 0.063.
  - Pre-treatment (t−1) average of asinh cumulative citations: 1.158 (corresponds to 1.43 untransformed cumulative citations).
  - Translated to percentage change using sinh: estimated increase of around 7.8% in citations to investing firms from the FDI destination country compared to non-destination countries.
  - 95% confidence interval for brownfield FDI implied percentage change: 5.7−10%.
- Greenfield FDI: coefficients statistically indistinguishable from brownfield, but with a lower pre-treatment average of 0.66 cumulative citations.
  - Implied percentage change in citations from greenfield FDI: 10.6%.
  - 95% confidence interval for greenfield FDI implied percentage change: 9−12.2%.

### Investors’ citations to host countries (baseline)
- Figures IIIc and IIId show increases in citations made to FDI host countries by investors of:
  - 10.8% and 13.4%, respectively.
- Note on greenfield pre-trend:
  - Figure IIId displays a small pre-trend indicating greenfield investors appear to select into destinations they are increasingly citing in the years leading up to entry.
  - Extrapolating this small pre-trend still implies a positive treatment effect, but the presence of the trend raises doubts on the validity of the parallel trend hypothesis for this specification and cautions against a causal interpretation.
  - The trend implied by coefficients 4 and 5 years before entry corresponds to 0.0067 asinh cumulative citations per year.
  - Subtracting this trend extrapolated to 5 periods after treatment implies an 11.8% increase in cumulative citations to host countries made by greenfield investors.

### Firm-country-industry specification (addressing pre-trends)
- Section 4.2 presents a firm-country-industry level specification that fully controls for pre-trends.
- There, the estimated increase in greenfield investors’ citations to patents in the targeted country-industry is 10.8%.
- This smaller coefficient is consistent with baseline findings because knowledge spillovers extend beyond directly targeted industries.

### US firms sample (Arora et al. (2021b) matched sample) and robustness
- Sample restriction to US firms with verified disambiguation and identification of patent ultimate owners (reflecting corporate ownership structures).
  - Patents owned by investors’ branches, subsidiaries or owned companies are attributed to the investor.
  - Investing firms in this sample will generally have more patents attributed to them compared to the world sample based on direct name matches.
- Computed effects (US sample):
  - Figures IVa and IVb display increases of 8.4% and 9.9% in citations received by investing firms, respectively.
  - Investors’ citations to host countries (Figures IVc and IVd): increases of 12.7% and 11.1%, respectively.
  - These numbers are within the ranges implied by the world-sample results and are not statistically different from Figure III estimates, given uncertainty.
- Pre-trends in US sample:
  - Slight pre-trend detected for greenfield FDI (Figure IVd); no violation of parallel trends detected for brownfield FDI.
  - Possible explanation: greenfield investors select destinations with accumulated relevant knowledge; brownfield FDI often targets firms in other primary SIC sectors.

### Patent reassignment concern and sensitivity
- Reassignment issue: In the Arora et al. (2021b) dataset, reassignment of patents to ultimate owners could mechanically increase citations attributed to investors if target-firm patents are reassigned to investors.
- Sensitivity test (Appendix A):
  - Restricting to citations of patents that were originally assigned to US firms does not significantly affect estimates of citations flowing from country c to firm i.
  - However, this restriction yields significantly lower measured knowledge spillovers in the opposite direction (citations made by investors to host countries), indicating part of the effect in Figures IVc and IVd is mechanical and driven by reassignment of destination-country patents to investors.
- Note: this reassignment issue is not present in the world sample, where patent owners correspond to original applicants and do not change over time.

### Overall summary of findings and caveats
- Baseline results suggest that FDI boosts knowledge diffusion both to and from FDI host countries with midpoints ranging from a 7.8% to a 12.7% increase in citations flowing between investors and host countries.
- Greenfield investments show larger implied percentage changes but exhibit detectable pre-trends that threaten a causal interpretation in the baseline specification.
- Additional specifications (firm-country-industry and robustness checks) reduce concerns about pre-trends and mechanical reassignment, yielding qualitatively similar but sometimes smaller estimated effects.
- The analysis highlights heterogeneity by FDI type (greenfield versus brownfield) and the importance of accounting for patent ownership/reassignment when interpreting investor-side citation increases.

*Source: 4.1 Main Result: Citation Flows Between Host Countries and Investing Firms (wpiea2024152-print-pdf)*

### 4.2    The Scope of Knowledge Diffusion: Untargeted Firms and Country-Industry

### 4.2    The Scope of Knowledge Diffusion: Untargeted Firms and Country-Industry

### Results overview
- Two additional specifications refine identification of knowledge spillovers and expand analysis scope: (i) excluding citations originating from firms acquired in brownfield FDI to address self-citation concerns, and (ii) analyzing citations at the country-industry (ISIC 2-digit) level to include country-industry-time and firm-industry-time fixed effects and to study technological linkages across industries.
- In the sample period, only 27.16% of M&A activities involve industries in the same primary SIC code.

### Dropping citations from target firms (robustness to self-citations)
- Purpose: exclude citations from acquired firms (name-matching to acquired targets or acquiring firms) to address mechanical self-citation by newly acquired subsidiaries. Applicable only to brownfield FDI (greenfield data do not include names of newly established entities).
- Samples: applied to both the world sample and the US sample; US sample provides greater ownership reliability.
- Key quantitative findings (Figure V results):
  - Increase in host country’s citations to brownfield investors’ patents:
    - World sample: 8.9%
    - US sample: 7.3%
    - These bounds cover baseline estimates of 7.8% and 8.4% for the same samples.
  - Investors’ citations to host countries:
    - World sample: 10.8% (the same as baseline)
    - US sample: 4.5% (significantly different from baseline; smaller size suggests about two thirds of the effect in Figure IVc is due to citations made by investors to acquired firms in destination countries)
- Controls and estimation details used in figures: firm-time and country-time fixed effects; three lags of the dependent variable in levels and its trend estimated using the change in 6 to 4 years before the event; stabilization period set to five periods.

### Firm-industry evidence: setup and specifications
- Industry-level specification (Equation (4)):
  - Outcome: y_icl,t+h − y_icl,t−1 where y_ilc,t is the asinh cumulative citations made to patents owned by firm i by patents assigned to country c’s patents for two-digit ISIC code l within a 5-year period including t, or vice versa.
  - Regressors: β_h T_icl,t + sum of three lag terms γ_hk y_icl,t−1−k + firm-industry-time fixed effects δ_hilt + country-industry-time fixed effects δ_hclt + error.
  - Treatment definition: first year when a US firm enters destination country c’s industry l.
  - Ability to include firm-industry-time and country-industry-time fixed effects addresses selection from target industries growing in country c or firms becoming more technologically relevant in sector l.
- Data mapping:
  - Greenfield FDI: fDi Markets descriptions mapped to ISIC 2-digit codes.
  - Brownfield FDI (M&A): only primary SIC sector available; less precise industry targeting information.
  - Therefore, primary presentation focuses on greenfield FDI; brownfield FDI discussed briefly with caveats.

### Knowledge input coefficients (treatment intensity by technological relatedness)
- Definition of knowledge input coefficient for industry l from industry k in country c (1990-2000):
  - a_c l←k ≡ Total citations made by industry l to k in country c between 1990-2000 / Total citations made by industry l in country c between 1990-2000
  - Properties: a_c l←k ranges between 0 and 1; sum across k equals 1.
  - Rationale: measures share of sector l’s citations directed to sector k; calculated using period 1990-2000 to avoid endogeneity with the main analysis.
- Treatment intensity with input coefficients:
  - T_icl,t ≡ sum_k D_ick,t × a_c l←k, where D_ick,t is the dummy for first year of FDI for firm-destination-industry triplet.
  - Interpretation: treatment is stronger for industries l that more intensely cite treated industries k; includes same-industry coefficient a_c l←l < 1 so directly targeted industries’ treatment is scaled accordingly.

### Sample selection and computational considerations
- Computational constraints prevent using all possible firm-country-industry-time combinations; analysis restricted to US sample and the top 30 innovative destinations (by share of worldwide patents of assignees’ patent offices).
- Resulting dataset scale:
  - Around 276 million observations.
  - Investments from US firms to the top 30 destinations across 98 ISIC 2-digit industries.
- Additional restriction: in the knowledge input coefficient specification, limited to firms that make only an investment to one target industry per country (encompassing 90% of total observations) to avoid arbitrary aggregation of input coefficients for multi-industry investors.

### Industry-level greenfield FDI results (Equation (4) estimates; Figure VI)
- Identification: inclusion of country-industry-time and firm-industry-time fixed effects; three lags and trend using change in 6 to 4 years before event; stabilization period five periods.
- Main quantitative findings:
  - Average increase in citations made by host industries to greenfield investors: 9%.
  - Investors’ citations to target industries: 10.8%.
  - These industry-level estimates are at the lower end of baseline country-level estimates, consistent with spillovers extending beyond targeted industries.
- Knowledge input coefficient interaction results:
  - Clear positive effects without pre-trends, indicating spillovers extend beyond directly targeted sectors.
  - Investments targeting an industry with input coefficient of 1 imply a 32.7% increase in citations made by related industries to greenfield investors (a_c l←k = 1 is a theoretical extreme not observed empirically).
  - Using average input coefficients interpretations:
    - Same-industry share a_l←l = 22.4% implies a 7.2% increase in citations for sectors directly targeted by FDI (this 7.2% is smaller than the 9% industry-level result but lies within the 6.6% lower bound implied by Figure VI confidence intervals).
    - Average treatment T_icl,t among industries with T_icl,t > 0 is 5.7% (on average industries connected to FDI target sectors direct 5.7% of their citations to those sectors), implying a 1.8% increase in citations made to investors by sectors affected directly or indirectly by FDI activity.
  - Complementary investor-side percent effects at first decimal identical to the above due to slightly higher coefficient and citations in asinh terms.
- Coverage of treated pairs under input-coefficient specification:
  - Directly targeted greenfield industry-country pairs: 3,487.
  - Treated industry-country pairs under knowledge input specification: 126,598 (3,487 equals 2.75% of 126,598).
- Interaction specification (T_icl,t × direct-target dummy):
  - Percent increase in citations for non-targeted industries with a_l←k = 1: 28.5%.
  - Corresponding number for targeted sectors: 22.7%.
  - Interpretation: spillovers (via input coefficients) can be larger in related non-targeted sectors in percent terms.

### Industry-level brownfield FDI results (limitations and findings)
- Industry assignment for brownfield FDI limited to primary industry of investors and targets, creating potential attenuation from mis-assigned treated industry.
- Findings when using acquiring firms’ industries:
  - Positive but not significant effects for citations from industries that are directly invested in to investor firms.
  - Imprecisely-estimated zero effects for investors’ citations to destination country-industries.
  - Likely explanation: acquiring firms’ primary industry may not match the actual industries targeted by the brownfield investment.
- Knowledge input coefficient specification for brownfield FDI:
  - Results similar to greenfield case; assigning T_icl,t > 0 via input coefficients helps recover spillovers because actual target industries rely on knowledge related to acquiring firms’ industry.
- When assigning direct treatment to industry l using target firm primary industry:
  - No effect on citations from investing firms to the target country-industry.
  - Positive effects on citations from the target country-industry to investing firms.
  - Interpretation: actual industry subject to FDI may be technologically related to the target firm’s primary industry but not coincide with it; target firm industry can receive indirect spillovers without generating spillovers back to investors.
- Additional note on industry mismatch:
  - The target and investors’ primary industry do not coincide in 75% of cases; this fuels measurement concerns for brownfield industry-level identification.
- Overall summary: subject to less precise industry identification in brownfield data, industry-level greenfield results generally extend to the brownfield context; related figures are available upon request.

*Source: IMF Working Paper — section 4.2, "The Scope of Knowledge Diffusion: Untargeted Firms and Country-Industry" (figures and specifications as presented in the source).*

### 4.3    Heterogeneity by Host Countries’ Absorptive Capacity

### 4.3    Heterogeneity by Host Countries’ Absorptive Capacity

### Empirical specification and absorptive-capacity measures
- Specification used (Eq. (6)) allows interaction of treatment with an absorptive-capacity term I_ic,t:
  - y_ic,t+h − y_ic,t−1 = β_h D_ic,t + β_I_h D_ic,t × I_ic,t + Σ_{k=1}^p γ_hk y_ic,t−k + η′ x_ic,t−1 + δ^h_it + δ^h_ct + ε^h_ic,t.
- Two proxies for host-country absorptive capacity:
  - Destination patent-stock dummy: I_ic,t = 1 if destination c belongs to the top 10% of destinations by the quantity of patents (PATSTAT).
    - Note: top five destinations in the sample are the U.S., China, Japan, Germany and Korea.
  - Technological similarity: cosine similarity between investors’ and countries’ patent-stock vectors across ISIC 2-digit industries, computed on patent stock accumulated in the previous ten years.
    - Cosine similarity multiplied by 10 for readability so I_ic,t ranges between 0 and 10.
    - For reference, I_ic,t averages 1.26 in the sample.

### Heterogeneity by destinations’ patent stocks (top 10% vs bottom 90%)
- Interaction term β_I_h is positive and significant: more citation flows arise in both directions when the FDI destination is among the top 10% patent producers.
- Countries’ citations to investors (Figure VII):
  - Brownfield FDI:
    - Bottom 90% destinations: citations increase by 5.1% after five years.
    - Top 10% destinations: citations increase by 12.3% after five years.
    - Being among top patent producing destinations more than doubles the estimated baseline effect in asinh terms.
  - Greenfield FDI:
    - Bottom 90% destinations: citations increase by 2.73%.
    - Top 10% destinations: citations increase by 21.3%.
- Investors’ citations to host countries (Figure VIII):
  - Brownfield investors → countries:
    - Bottom 90% destinations: citations increase by 6.8%.
    - Top 10% destinations: citations increase by 17.6%.
  - Greenfield investors → countries:
    - Bottom 90% destinations: citations increase by 4.9%.
    - Top 10% destinations: citations increase by 24.5%.
- Internal-validity note:
  - Interaction compares same firm across different destinations and firms investing in top 10% versus others.
  - Authors repeat specification restricting to multi-destination firms; results qualitatively unchanged but noisier.

### Heterogeneity by technological similarity (cosine similarity)
- Measure: cosine similarity of investing firms’ and host countries’ patent stocks (ISIC 2-digit), averaged over past ten years; multiplied by 10 so typical I_ic,t values: 0–10; sample mean 1.26.
- Main patterns (Figures IX and X):
  - Higher technological similarity generates larger citation flows.
  - Countries benefit from knowledge spillovers even when technological similarity is very low (upper panels of Figure IX).
  - For investors’ citations to hosts, when destination patent stock similarity is 0, there are no investors’ citations from greenfield FDI investors.
  - Technological similarity plays a lesser role for brownfield FDI than for greenfield FDI:
    - Moving from the 25th to the 75th percentile of similarity (from 0.33 to 1.66) increases citations generated by brownfield FDI by about 32%.
    - The same move more than doubles the effect of greenfield FDI.
- Quantitative examples:
  - For greenfield-related host-country citations:
    - Citation effects reported: 6.7% at the 25th percentile and 9.2% at the 75th percentile (context: greenfield FDI citation effects).
  - For investors’ citations to host countries (greenfield context), no increase when patent-stock similarity is 0.
  - For the specification summarized in the text:
    - Citations increase by 4% at the 25th percentile of similarity and 11.1% at the 75th percentile (context described in the text for a reported figure).
    - For brownfield investments reported in Figure Xa, increases are 12% (25th) and 10% (75th).
  - Note on interpretation: percentage effects may be slightly lower in higher-similarity countries because pre-treatment citation levels are higher there; absolute changes remain larger at higher similarity.

### Other results and robustness
- Patent creation:
  - No conclusive evidence on the effects of FDI on patent creation due to identification challenges.
  - Attempts with brownfield FDI on target firms produced unreliable results:
    - Effects appear negative for granted patents and positive for triadic patents.
    - Possible mechanical explanation: target firms may cease separate operations after acquisition.
  - Results on patent-creation exercises available upon request.
- Robustness checks (baseline results remain robust under):
  - (1) Expanding destination set to all those in PATSTAT.
  - (2) Limiting forward citations to five years after patent application.
  - (3) Increasing stabilization lags L to 12.
  - (4) Using log(1 + x) instead of asinh(x) for cumulative citations.
  - (5) Using only citations between triadic patents.
  - (6) Using each investment (not just first FDI) as a separate event.
  - (7) Grouping greenfield and brownfield FDI into a single “FDI investment” variable.
  - With few exceptions, coefficients and implied percentage changes are not statistically different from baseline.
  - For the US sample using only patents originally assigned to US firms: results for citations made by US firms to destination countries decrease in magnitude but remain positive and statistically significant.
  - Additional details and figures are reported in Appendix A.

### Key summary findings and implications (from section conclusion)
- Average estimated citation changes after first investor entry:
  - Destinations’ citations to investors: increase by 7.8%-10.6% on average, depending on investment type.
  - Investors’ citations to FDI hosts: increase by 4.5%-13%.
- Knowledge spillovers extend beyond targeted firms and industries to benefit other sectors in destination countries.
- Host-country absorptive capacity matters:
  - Hosts with larger pre-existing patent stocks and technologies more similar to investors benefit more from both brownfield and greenfield investments.
- Policy and research implications:
  - Sizable knowledge flows induced by FDI suggest potential productivity and growth losses if investment fragmentation occurs amid geopolitical tensions.
  - Destinations’ capacity to reap FDI benefits is critical; further investigation needed into channels of knowledge transmission and heterogeneity by investment type (brownfield vs greenfield).
  - Future work: classification of FDI motives and characteristics using text analysis; investigation of diffusion of non-patent organizational knowledge (know-how); whether acquired firms/industries increase overall patenting; role of FDI in catalyzing new economic sectors.

*Source: Section 4.3 from the provided IMF working paper content.*

### References

### wpiea2024152-print-pdf - References

### References
- Aghion, P., A. Bergeaud, T. Gigout, M. Lequien, and M. Melitz (2023): “Exporting ideas: Knowledge flows from expanding trade in goods.”
- Aitken, B. J. and A. E. Harrison (1999): “Do domestic firms benefit from direct foreign investment? Evidence from Venezuela,” American economic review, 89, 605–618.
- Aiyar, S., D. Malacrino, and A. Presbitero (2023): “Investing in friends: The role of geopolitical alignment in fdi flows,” Tech. rep., CEPR Discussion Paper.
- Akcigit, U., S. T. Ates, J. Lerner, R. R. Townsend, and Y. Zhestkova (2024): “Fencing off Silicon Valley: Cross-border venture capital and technology spillovers,” Journal of Monetary Economics, 141, 14–39.
- Akcigit, U. and M. Melitz (2022): “International trade and innovation,” in Handbook of International Economics, Elsevier, vol. 5, 377–404.
- Alfaro, L. (2017): “Multinational Activity in Emerging Markets: How and When Does Foreign Direct Investment Promote Growth?” in Geography, Location, and Strategy, Emerald Publishing Limited, vol. 36 of Advances in Strategic Management, 429–462.
- Alfaro, L., A. Chanda, S. Kalemli-Ozcan, and S. Sayek (2004): “FDI and economic growth: the role of local financial markets,” Journal of international economics, 64, 89–112.
- Antràs, P. and S. R. Yeaple (2014): “Multinational firms and the structure of international trade,” Handbook of international economics, 4, 55–130.
- Arora, A., S. Belenzon, and L. Sheer (2021a): “Knowledge spillovers and corporate investment in scientific research,” American Economic Review, 111, 871–898.
- Arora, A., S. Belenzon, and L. Sheer (2021b): “Matching patents to Compustat firms, 1980–2015: Dynamic reassignment, name changes, and ownership structures,” Research Policy, 50, 104217.
- Autor, D., D. Dorn, G. H. Hanson, G. Pisano, and P. Shu (2020): “Foreign competition and domestic innovation: Evidence from US patents,” American Economic Review: Insights, 2, 357–374.
- Azoulay, P., C. Fons-Rosen, and J. S. G. Zivin (2019): “Does science advance one funeral at a time?” American Economic Review, 109, 2889–2920.
- Belz, T., L. A. Robinson, M. Ruf, and C. Steffens (2013): “Tax avoidance as a driver of mergers and acquisitions,” Available at SSRN 2371706.
- Bena, J. and K. Li (2014): “Corporate innovations and mergers and acquisitions,” The Journal of Finance, 69, 1923–1960.
- Bergant, K., P. Mishra, and R. Rajan (2023): “Cross-border Spillovers: How US Financial Conditions affect M&As Around the World,” Tech. rep., National Bureau of Economic Research.
- Bernard, A. B., J. B. Jensen, and P. K. Schott (2009): “Importers, exporters and multi-nationals: a portrait of firms in the US that trade goods,” in Producer dynamics: New evidence from micro data, University of Chicago Press, 513–552.
- Blalock, G. and P. J. Gertler (2008): “Welfare gains from foreign direct investment through technology transfer to local suppliers,” Journal of international Economics, 74, 402–421.
- Bollaert, H. and M. Delanghe (2015): “Securities Data Company and Zephyr, data sources for M&A research,” Journal of Corporate Finance, 33, 85–100.
- Borensztein, E., J. De Gregorio, and J.-W. Lee (1998): “How does foreign direct investment affect economic growth?” Journal of international Economics, 45, 115–135.
- Branstetter, L. (2006): “Is foreign direct investment a channel of knowledge spillovers? Evidence from Japan’s FDI in the United States,” Journal of International economics, 68, 325–344.
- Chen, J. and J. Roth (2024): “Logs with zeros? Some problems and solutions,” The Quarterly Journal of Economics, 139, 891–936.
- Chen, S.-F. S. and M. Zeng (2004): “Japanese investors’ choice of acquisitions vs. startups in the US: The role of reputation barriers and advertising outlays,” International Journal of Research in Marketing, 21, 123–136.
- Chen, Y., H. Jiang, Y. Liang, and S. Pan (2022): “The impact of foreign direct investment on innovation: evidence from patent filings and citations in China,” Journal of Comparative Economics, 50, 917–945.
- Cunningham, C., F. Ederer, and S. Ma (2021): “Killer acquisitions,” Journal of Political Economy, 129, 649–702.
- de Chaisemartin, C. and X. D’Haultfœuille (2020): “Two-Way Fixed Effects Estimators with Heterogeneous Treatment Effects,” American Economic Review, 110, 2964–2996.
- de Rassenfosse, G., H. Dernis, and G. Boedt (2014): “An Introduction to the Patstat Database with Example Queries,” Australian Economic Review, 47, 395–408.
- Deng, L., Y. Lu, and Y. Tang (2024): “Does FDI increase product innovation of domestic firms? Evidence from China,” Journal of Economic Behavior & Organization, 222, 1–24.
- Dikova, D. and K. Brouthers (2016): “International establishment mode choice: Past, present and future,” Management International Review, 56, 489–530.
- Dube, A., D. Girardi, and A. M. Taylor (2023): “A Local Projections Approach to Difference-in-Differences Event Studies,” NBER Working Paper 31184.
- Erel, I., Y. Jang, and M. S. Weisbach (2022): “Cross-border mergers and acquisitions,” Tech. rep., National Bureau of Economic Research.
- Glass, A. J. and K. Saggi (1998): “International technology transfer and the technology gap,” Journal of development economics, 55, 369–398.
- Globerman, S., A. Kokko, and F. Sjöholm (2000): “International technology diffusion: evidence from Swedish patent data,” Kyklos, 53, 17–38.
- Goldbach, S., A. J. Nagengast, E. Steinmüller, and G. Wamser (2019): “The effect of investing abroad on investment at home: On the role of technology, tax savings, and internal capital markets,” Journal of International Economics, 116, 58–73.
- Goodman-Bacon, A. (2021): “Difference-in-Differences with Variation in Treatment Timing,” Journal of Econometrics, 225, 254–277.
- Gopinath, G., P.-O. Gourinchas, A. Pescatori, and P. Topalova (2024): “Changing Global Linkages: A New Cold War?” IMF Working Paper, WP/24/76.
- Griffith, R., R. Harrison, and J. Van Reenen (2006): “How special is the special relationship? Using the impact of US R&D spillovers on UK firms as a test of technology sourcing,” American Economic Review, 96, 1859–1875.
- Harrison, A. and A. Rodríguez-Clare (2010): “Trade, foreign investment, and industrial policy for developing countries,” Handbook of development economics, 5, 4039–4214.
- Haskel, J. E., S. C. Pereira, and M. J. Slaughter (2007): “Does inward foreign direct investment boost the productivity of domestic firms?” The review of economics and statistics, 89, 482–496.
- IMF (2018): “Chapter 4: Is Productivity Growth Shared in a Globalized Economy?” in World Economic Outlook, April 2018: Cyclical Upswing, Structural Change.
- IMF (2023): “Chapter 4: Geoeconomic Fragmentation and Foreign Direct Investment,” in World Economic Outlook, April 2023: A Rocky Recovery.
- Jaffe, A. B., M. Trajtenberg, and R. Henderson (1993): “Geographic localization of knowledge spillovers as evidenced by patent citations,” the Quarterly journal of Economics, 108, 577–598.
- Javorcik, B. S. (2004): “Does foreign direct investment increase the productivity of domestic firms? In search of spillovers through backward linkages,” American economic review, 94, 605–627.
- Javorcik, B. S., A. Lo Turco, and D. Maggioni (2018): “New and improved: does FDI boost production complexity in host countries?” The Economic Journal, 128, 2507–2537.
- Jiang, K., W. Keller, L. D. Qiu, and W. Ridley (2018): “International joint ventures and internal vs. external technology transfer: Evidence from China,” Tech. rep., National Bureau of Economic Research.
- Jinji, N., X. Zhang, and S. Haruna (2015): “Trade patterns and international technology spillovers: evidence from patent citations,” Review of World Economics, 151, 635–658.
- Keller, W. (1996): “Absorptive capacity: On the creation and acquisition of technology in development,” Journal of development economics, 49, 199–227.
- Keller, W. (2021): “Knowledge spillovers, trade, and FDI,” Tech. rep., National Bureau of Economic Research.
- Keller, W. and S. R. Yeaple (2009): “Multinational enterprises, international trade, and productivity growth: firm-level evidence from the United States,” The review of economics and statistics, 91, 821–831.
- LaBelle, J., I. Martinez-Zarzoso, A. M. Santacreu, and Y. V. Yotov (2023): “Cross-border Patenting, Globalization, and Development.”
- Lybbert, T. J. and N. J. Zolas (2014): “Getting patents and economic data to speak to each other: An ‘Algorithmic Links with Probabilities’ approach for joint analyses of patenting and economic activity,” Research Policy, 43, 530–542.
- Maurseth, P. B. and B. Verspagen (2002): “Knowledge spillovers in Europe: a patent citations analysis,” Scandinavian Journal of Economics, 104, 531–545.
- Moretti, E. (2021): “The effect of high-tech clusters on the productivity of top inventors,” American Economic Review, 111, 3328–3375.
- Navaretti, G. B. and A. J. Venables (2004): Multinational firms in the world economy, Princeton University Press.
- Neary, J. P. (2007): “Cross-border mergers as instruments of comparative advantage,” The Review of Economic Studies, 74, 1229–1257.
- Ni, B., H. Kato, and Y. Liu (2021): “Does It Matter Where You Invest? The Impact of FDI on Domestic Job Creation and Destruction,” The Impact of FDI on Domestic Job Creation and Destruction (January 24, 2021).
- Phillips, G. M. and A. Zhdanov (2013): “R&D and the incentives from merger and acquisition activity,” The Review of Financial Studies, 26, 34–78.
- Rodriguez-Clare, A. (1996): “Multinationals, linkages, and economic development,” The American economic review, 852–873.
- Silva, J. S. and S. Tenreyro (2006): “The log of gravity,” The Review of Economics and statistics, 88, 641–658.
- Toews, G. and P.-L. Vézina (2022): “Resource discoveries, FDI bonanzas, and local multipliers: Evidence from Mozambique,” Review of Economics and Statistics, 104, 1046–1058.

### Appendix: Robustness Results
- Baseline percent-change findings (as reported, with 95% C.I. bounds in parentheses):
  - (a) Citations to brownfield investors increase on average by 7.8% (5.7−10%).
  - (b) Citations to greenfield investors increase on average by 10.6% (9−12.2%).
  - (c) Citations to countries made by brownfield investors increase by 10.8% (8.4−13.3%).
  - (d) Citations to countries made by greenfield investors increase by 13.4% (11.6−15.2%).
- Robustness: Including All Available Countries
  - Midpoint increases: (a) 7.8%; (b) 9.9%; (c) 11.3%; (d) 13.1%.
- Robustness: Fixing the Citation Window to 5 Years (sample restricted to 2003–2015; PATSTAT Spring 2022)
  - Percentage increases: (a) 11.3%; (b) 12%; (c) 15.1%; (d) 7.7%.
  - Note: (a) and (c) exceed baseline upper 95% C.I. bounds; (d) is below baseline.
- Robustness: Increasing L to 12 years (stabilization periods)
  - Percentage increases: (a) 5.6%; (b) 12.5%; (c) 10.8%; (d) 12.5%.
  - Note: (a) sits just below baseline lower bound of 5.7%; (b) slightly above baseline 12.2%.
- Robustness: Using log(1 + x) Transformation (instead of asinh)
  - Increases in citations: (a) 9.4% ; (b) 12.7%; (c) 13.1%; (d) 16.2%.
  - Note: Midpoints always larger than baseline; panels (b) and (d) significantly so.
- Robustness: Using Triadic Patents Only (registered at USPTO, JPO, EPO)
  - Citation increases: (a) 6.6%; (b) 9.5%; (c) 10%; (d) 9.5%.
  - Note: Midpoints lower than baseline; significantly lower only for (d).
- Robustness: Using Each Investment as a Separate Event
  - Number of events increases: about 12% more for brownfield and about 20% more for greenfield.
  - Percentage increases: (a) 6.8%; (b) 9.9%; (c) 9.5%; (d) 13.2%.
  - Note: Estimates not significantly different from baseline.
- Robustness: Grouping Greenfield and Brownfield FDI (treatment = any FDI; require no other FDI in previous five years; each investment separate event)
  - Number of events: 29,573.
  - Findings: citations from destinations to FDI investors increase by 9.4%; investors’ citations to target countries rise 12.9%.
  - Note: These align closely with simple averages of (a)/(b) = 9.1% and (c)/(d) = 12.1%, and lie within baseline 95% C.I. bounds.
- Robustness: Using Only Patents Originally Assigned to US Firms (US sample)
  - Percentage increases in lifetime citations: (a) 5.7%; (b) 9.1%; (c) 3.5%; (d) 5.4%.
  - Comparison (Figure IV with C.I. bounds): (a) 8.4% (4.9−11.9%); (b) 10% (7.2−12.7%); (c) 12.7% (7−18.6%); (d) 11.1% (6.2−16.1%).
  - Note: Results on citations flowing from FDI investors to destination countries ((c) and (d)) are substantially affected by restricting to original US owners, suggesting part of the baseline effect arises from mechanical reassignment of patents to investing U.S. firms.

*Knowledge Diffusion Through FDI: Worldwide Firm-Level Evidence — Working Paper No. WP/2024/152*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024152-print-pdf.pdf_