Pre-registration
The cumulative EU digital-sector regulatory stack — GDPR (2018), Digital Markets Act (2022), Digital Services Act (2022), AI Act (2024) — has imposed a measurable fixed-compliance-cost burden that falls disproportionately on scale of EU-headquartered digital firms relative to US counterparts. Over 2018-2024, EU-headquartered digital firms show lower revenue-per-firm, lower VC-funding-per-firm, lower mean-employee count, and a lower rate of scale-up (defined as growth from <100 to >500 employees) than comparable-stage US firms. The effect is hypothesised to operate through the fixed-cost channel (compliance cost is lumpy and amortised over revenue base; smaller market means thinner spreading). The hypothesis does NOT claim the regulations produce net welfare loss (GDPR's consumer-privacy externality is explicitly outside the outcome set) — only that firm-scale is differentially smaller on the measured dimensions.
Falsification criterion — what would disprove this
This hypothesis is considered falsified if:
Not supported if (a) β_eu_post_2018 is non-negative or not statistically significant at p<0.10 on at least two of the four outcome measures, OR (b) the effect vanishes after controlling for capital-market depth and VC-raised-as-share-of-GDP, OR (c) placebo at pre-2018 fake date shows a similar-magnitude "effect" (indicating pre-existing trend rather than GDPR), OR (d) the scale-up rate difference is driven by US superstar firms rather than EU-firm scale-up deficit (decomposable via medians vs means). Report honestly if the scale-gap exists but cannot be attributed to regulation specifically — this is a real and legitimate null.
formal test & threshold
test: eu_digital_regulation_firm_scale_did threshold: β_eu_post_2018 < -0.05 log points on log VC funding per firm AND β_eu_post_2018 < -0.02 on scaleup_rate (at p<0.10) AND placebo pre-2018 |t| < 1.65 AND effects survive capital-market-depth control
Method
- Template
panel_fe- Fixed effects
country, year- Clustering
country- Sample
- 15 countries · 2014 – 2024
- Evidence type
- associational
Two-way FE panel with USA, CAN, and post-2020 GBR as non-EU control. β_eu_post_2018 identifies the EU-member post-GDPR average shift on the four outcome measures. Secondary: event-study around GDPR enforcement date (May 2018), DMA entry into application (2023), and AI Act enforcement phases. Separate coefficient per episode; cumulative burden tested as sum. Identification is associational rather than causal: multiple treatments coincide with other large shocks (COVID, gas-price shock, US Big Tech concentration continuing), and the pure regulation-compliance channel cannot be cleanly separated from co-occurring forces. Placebo at pre-2018 fake dates tested. Known limitations: (1) Crunchbase / PitchBook firm-level data has uneven country coverage; EU coverage weaker than US. Measurement bias could under- or overstate effects in unknown directions. (2) Capital-market depth is a confound with known importance — EU VC markets are thinner independently of regulation. Controlling for VC raised as share of GDP partly addresses, but collinearity with treatment is real. (3) Big Tech concentration in the US is partly a tax and antitrust story independent of EU regulation. The comparison is not "EU-regulated vs counterfactual EU-unregulated" but "EU-regulated vs US-different-bundle."
Data
| Variable | Source | Transform |
|---|---|---|
log_mean_digital_firm_revenue outcome | constructed:Crunchbase / PitchBook firm-level digital-sector revenue, aggregated to country-year mean (log). Firm-level data-gated ftier 5 | log |
log_vc_funding_per_active_firm outcome | constructed:total VC dollars invested in digital-sector firms per country-year / number of active digital firms. Crunchbase fetcher tier 5 | log |
scaleup_rate outcome | constructed:fraction of digital firms that transitioned from <100 to >500 employees within the country-year window. Crunchbase / Pittier 5 | level |
digital_sector_employment_share outcome | constructed:share of total employment in ISIC J (information and communication). OECD STAN or ILO where available.tier 5 | level |
eu_post_2018_gdpr_dummy treatment | constructed:indicator = 1 for EU member states in year 2018 onwards. USA, CAN, GBR (post-2020) = control.tier 5 | indicator |
eu_post_2022_dma_dsa_dummy treatment | constructed:indicator for EU member states in 2022 onwards, incremental cumulative burden.tier 5 | indicator |
eu_post_2024_ai_act_dummy treatment | constructed:indicator for EU member states in 2024 onwards; 2024 effect will be shallow given recency.tier 5 | indicator |
log_population control | world_bank_wdi:SP.POP.TOTLtier 2 | log |
oecd_pmr_digital_subindex control | oecd_pmr:OECD.ECO.GCRDtier 4 | level |
log_gdp_pc_ppp control | world_bank_wdi:NY.GDP.PCAP.PP.KDtier 2 | log |
capital_market_depth_proxy control | constructed:domestic VC dollars raised as share of GDP (Invest Europe + NVCA). Fetcher pending.tier 5 | level |
● ready · ● pending · ● reconstruct-needed
Detailed result card
Result card — gdpr_digital_sector_firm_scale_effect
Verdict: INCONCLUSIVE_DATA_PENDING — no outcome variable loaded; missing: ['constructed: Crunchbase / PitchBook firm-level digital-sector revenue, aggregated to country-year mean (log). Firm-level data-gated fetcher pending.', 'constructed: total VC dollars invested in digital-sector firms per country-year / number of active digital firms. Crunchbase fetcher pending.', 'constructed: fraction of digital firms that transitioned from <100 to >500 employees within the country-year window. Crunchbase / PitchBook.', 'constructed: share of total employment in ISIC J (information and communication). OECD STAN or ILO where available.']
Pre-registration
- Claim: The cumulative EU digital-sector regulatory stack — GDPR (2018), Digital Markets Act (2022), Digital Services Act (2022), AI Act (2024) — has imposed a measurable fixed-compliance-cost burden that falls disproportionately on scale of EU-headquartered digital firms relative to US counterparts. Over 2018-2024, EU-headquartered digital firms show lower revenue-per-firm, lower VC-funding-per-firm, lower mean-employee count, and a lower rate of scale-up (defined as growth from <100 to >500 employees) than comparable-stage US firms. The effect is hypothesised to operate through the fixed-cost channel (compliance cost is lumpy and amortised over revenue base; smaller market means thinner spreading). The hypothesis does NOT claim the regulations produce net welfare loss (GDPR's consumer-privacy externality is explicitly outside the outcome set) — only that firm-scale is differentially smaller on the measured dimensions.
- Falsification rule: Not supported if (a) β_eu_post_2018 is non-negative or not statistically significant at p<0.10 on at least two of the four outcome measures, OR (b) the effect vanishes after controlling for capital-market depth and VC-raised-as-share-of-GDP, OR (c) placebo at pre-2018 fake date shows a similar-magnitude "effect" (indicating pre-existing trend rather than GDPR), OR (d) the scale-up rate difference is driven by US superstar firms rather than EU-firm scale-up deficit (decomposable via medians vs means). Report honestly if the scale-gap exists but cannot be attributed to regulation specifically — this is a real and legitimate null.
- Falsification test: eu_digital_regulation_firm_scale_did
Estimate
- Error: no outcome variable loaded; missing: ['constructed: Crunchbase / PitchBook firm-level digital-sector revenue, aggregated to country-year mean (log). Firm-level data-gated fetcher pending.', 'constructed: total VC dollars invested in digital-sector firms per country-year / number of active digital firms. Crunchbase fetcher pending.', 'constructed: fraction of digital firms that transitioned from <100 to >500 employees within the country-year window. Crunchbase / PitchBook.', 'constructed: share of total employment in ISIC J (information and communication). OECD STAN or ILO where available.']
Variables resolved
constructed: indicator = 1 for EU member states in year 2018 onwards. USA, CAN, GBR (post-2020) = control.→ eu_post_2018_gdpr_dummy (treatment, publisher=constructed, n=165)constructed: indicator for EU member states in 2022 onwards, incremental cumulative burden.→ eu_post_2022_dma_dsa_dummy (treatment, publisher=constructed, n=165)constructed: indicator for EU member states in 2024 onwards; 2024 effect will be shallow given recency.→ eu_post_2024_ai_act_dummy (treatment, publisher=constructed, n=165)world_bank_wdi:SP.POP.TOTL→ log_population (controls, publisher=world_bank_wdi, n=14447)oecd_pmr:OECD.ECO.GCRD,DSD_PMR@DF_PMR,1.2→ oecd_pmr_digital_subindex (controls, publisher=oecd_pmr, n=105)world_bank_wdi:NY.GDP.PCAP.PP.KD→ log_gdp_pc_ppp (controls, publisher=world_bank_wdi, n=8325)
Variables missing data
constructed: Crunchbase / PitchBook firm-level digital-sector revenue, aggregated to country-year mean (log). Firm-level data-gated fetcher pending.(outcome, name=log_mean_digital_firm_revenue) — vintage not on diskconstructed: total VC dollars invested in digital-sector firms per country-year / number of active digital firms. Crunchbase fetcher pending.(outcome, name=log_vc_funding_per_active_firm) — vintage not on diskconstructed: fraction of digital firms that transitioned from <100 to >500 employees within the country-year window. Crunchbase / PitchBook.(outcome, name=scaleup_rate) — vintage not on diskconstructed: share of total employment in ISIC J (information and communication). OECD STAN or ILO where available.(outcome, name=digital_sector_employment_share) — vintage not on diskconstructed: domestic VC dollars raised as share of GDP (Invest Europe + NVCA). Fetcher pending.(controls, name=capital_market_depth_proxy) — vintage not on disk
Generated by scripts/run_panel_fe.py at 2026-06-29T17:54:29+00:00
Strongest opposing argument
Every hypothesis ships with its charitable opposing argument. The framework earns credibility by handling objections at their strongest, not weakest.
Notes
Data readiness: - Crunchbase / PitchBook firm-level — commercial, data-gated, fetcher pending and possibly requires partnership - OECD PMR digital subindex — ready - OECD STAN ICT sector VA — fetcher pending - Invest Europe + NVCA VC totals — fetcher pending - WDI, IMF — ready This hypothesis is data-gated on firm-level commercial sources. A weaker v1 using only sector-aggregate outcomes (ICT VA, ICT employment, aggregate VC inflows) is runnable now; v1.1 with firm-level outcomes when commercial data access lands.