IESET.
Hypotheses·growth·industrial_policy_developmentalist_states_growth

Developmentalist East Asian states (South Korea, Taiwan, Singapore, China) pursuing active industrial policy — export-discipline, selective credit, state-directed FDI screening, targeted sector promotion — achieved higher long-run real per-capita GDP growth over 1960-2019 than otherwise-comparable countries starting at similar income levels in 1960.

The comparable-control set is constructed from Maddison 1960 per-capita GDP neighbours plus similar initial human-capital and geographic endowment, drawn from Latin America (ARG, BRA, MEX, CHL), South Asia (IND, PAK, LKA), Southeast Asia non-developmentalist (PHL, THA in the earlier period), and Africa (GHA, EGY). The hypothesis is that the developmentalist subset's long-run growth trajectory exceeds the control group by a margin not explained by controls for initial human capital, geography, and institutional quality alone.

SUPPORTEDengine/runs/industrial_policy_developmentalist_states_growth

SUPPORTED — avg ATT across 4 developmentalist cases (KOR/TWN/SGP/CHN) is +1.088 log-points at 40-yr horizon (~+197%). 4/4 cases above the 30 log-point threshold. Mean per-case placebo rank-p = 0.20. Polity-restricted attenuation check NOT RUN (Polity5 vintage not in repo); the polity-positive subset attenuation gate is DEFERRED.

confidence cueThis is a clear pass for the claim as written. It still applies only to this sample, period, and method.

policy briefNeeds review

In ordinary language

Over a long period, do more market-oriented institutions translate into higher income or productivity, once the comparison looks beyond a single success story?

plain answer

The data clearly moved in the predicted direction. avg ATT across 4 developmentalist cases (KOR/TWN/SGP/CHN) is +1.088 log-points at 40-yr horizon (~+197%).

why it matters

Growth claims can look convincing in single success stories. This test asks whether the pattern survives a broader comparison.

how the test works

It compares 19 country or place units from 1960 to 2019, using a synth did design.

what was measured
What changed
  • Developmentalist state category
What we checked
  • Real income per capita cost-of-living adjusted
  • Manufacturing value added share
  • Export sophistication index
what this does not prove

A single test is not the whole truth. It narrows the claim under a specific sample, time period, and method. Strong policy conclusions need the pattern to survive nearby tests, alternative data, and serious objections.

verification

1 input datasets, 0 unresolved missing series, provenance status: partial provenance.

Results

engine/runs/industrial_policy_developmentalist_states_growth
Loading chart…

Who has skin in the game — schools predicting on this

17 schools list this hypothesis as a test of their position. The chips below are school-level scoreboard outcomes, not a second hypothesis verdict.

hypothesis verdict vs scoreboard outcome

The banner verdict judges this hypothesis as written. The scoreboard asks whether each school's polarity-corrected prediction was right. Raw status is not a school win: SUPPORTED supports schools that needed SUPPORTED, but refutes schools that needed REFUTED.

Austrian Business Cycle Theory (ABCT)wins if REFUTEDschool refutedChicago School Monetarismwins if REFUTEDschool refutedClassical Liberalwins if REFUTEDschool refutedDegrowth / Ecological Macroeconomicswins if MIXEDpartial/no-callDemocratic Socialism (Bernsteinian revisionism + modern)wins if SUPPORTEDschool supportedStructuralist / Developmentalistwins if SUPPORTEDschool supportedEco-Socialism / Red-Greenwins if SUPPORTEDschool supportedEmpirical Pragmatist (the framework's house position)wins if MIXEDpartial/no-callNew Institutional Economicswins if SUPPORTEDschool supportedMarket Socialism (Lange-Lerner tradition)wins if SUPPORTEDschool supportedMarxian Political Economywins if SUPPORTEDschool supportedMarxism-Leninism (central-planning implementation)wins if MIXEDpartial/no-callModern Monetary Theorywins if SUPPORTEDschool supportedNew Keynesian (mainstream synthesis)wins if MIXEDpartial/no-callOrdoliberalism (Freiburg School)wins if MIXEDpartial/no-callPost-Keynesianwins if SUPPORTEDschool supportedSocial Democraticwins if SUPPORTEDschool supported

Pre-registration

pre-registered
first-spec commit 4c8ce8e · 2026-07-18T22:11:21Z

Developmentalist East Asian states (South Korea, Taiwan, Singapore, China) pursuing active industrial policy — export-discipline, selective credit, state-directed FDI screening, targeted sector promotion — achieved higher long-run real per-capita GDP growth over 1960-2019 than otherwise-comparable countries starting at similar income levels in 1960. The comparable-control set is constructed from Maddison 1960 per-capita GDP neighbours plus similar initial human-capital and geographic endowment, drawn from Latin America (ARG, BRA, MEX, CHL), South Asia (IND, PAK, LKA), Southeast Asia non-developmentalist (PHL, THA in the earlier period), and Africa (GHA, EGY). The hypothesis is that the developmentalist subset's long-run growth trajectory exceeds the control group by a margin not explained by controls for initial human capital, geography, and institutional quality alone.

Falsification criterion — what would disprove this

set before the run · honoured after

This hypothesis is considered falsified if:

Not supported if (a) the synthetic-DiD ATT on log real per-capita GDP averaged over the four developmentalist cases is not positive and outside the placebo distribution at p<0.10, OR (b) fewer than 3 of 4 individual cases show a positive ATT of at least 30 log points at the 40-year post-treatment horizon (the expected magnitude is 50-150 log points based on the Maddison trajectories), OR (c) the ATT attenuates by more than 50% when the comparable- initial-conditions matching is restricted to countries with Polity2 > 0 in the pre-treatment year (removing the authoritarian developmentalist / democratic-control confounder).

formal test & threshold
test:      synth_did_plus_individual_case_magnitudes
threshold: avg_ATT(log_gdp_pc, h=40y) > 0 at p<0.10 in placebo distribution AND 3 of 4 cases ATT >= 30 log points AND ATT_attenuation_restricted_to_polity_positive < 50%

Method

Template
synth_did
Clustering
country
Sample
19 countries · 1960 – 2019
Evidence type
causal

Synthetic DiD on each of the four developmentalist cases (KOR 1961, TWN 1958 or 1960, SGP 1965, CHN 1978) against a donor pool of the Latin American, South Asian, Southeast Asian non-developmentalist, and African controls. Pre-treatment fit on initial GDP, human capital, geography, and institutional quality. ATT averaged across the four cases with uncertainty intervals from the Abadie- Diamond-Hainmueller placebo distribution. Robustness: panel FE on the full sample with developmentalist-category dummy reported as v2 companion.

Data

VariableSourceTransform
real_gdp_per_capita_ppp
outcome
maddison:real_gdp_pctier 3
world_bank_wdi:NY.GDP.PCAP.KDtier 2
log
manufacturing_value_added_share
outcome
world_bank_wdi:NV.IND.MANF.ZStier 2
level_pct_of_gdp
export_sophistication_index
outcome
owid:economic-complexity-indextier 2
level_eci
developmentalist_state_category
treatment
constructed:binary indicator per country per epochtier 5
binary
initial_human_capital_1960
control
owid:mean-years-of-schooling-long-run-1870tier 2
level_at_1960
initial_gdp_per_capita_1960
control
maddison:real_gdp_pctier 3
log_level_at_1960
institutional_quality_pretreatment
control
polity5:polity2tier 4
level_at_treatment_start
geographic_latitude_abs
control
constructed:abs(capital_latitude) from standard geographic tabletier 5
level
primary_export_share
control
world_bank_wdi:TX.VAL.MRCH.R6.ZStier 2
level_pct

● ready  · ● pending  · ● reconstruct-needed

Detailed result card

Result card — Industrial policy developmentalist states growth

Verdict: SUPPORTED — avg ATT across 4 developmentalist cases (KOR/TWN/SGP/CHN) is +1.088 log-points at 40-yr horizon (~+197%). 4/4 cases above the 30 log-point threshold. Mean per-case placebo rank-p = 0.20. Polity-restricted attenuation check NOT RUN (Polity5 vintage not in repo); the polity-positive subset attenuation gate is DEFERRED.

Per-case synthetic control on log real GDP per capita (Maddison Project mpd2020). Treated cases: {'KOR': 1961, 'TWN': 1960, 'SGP': 1965, 'CHN': 1978}. Donor pool: ARG, BRA, MEX, CHL, IND, PAK, LKA, PHL, THA, GHA, EGY, TUR, IDN. Horizon: 40 years post-treatment, censored to 2018 (Maddison end).

Per-case ATT and placebo rank

| Case | t-year | Pre-RMSPE | Post-RMSPE | Ratio | ATT @ horizon | Placebo rank | p | |---|---:|---:|---:|---:|---:|:---:|---:| | KOR | 1961 | 0.018 | 0.886 | 49.74 | +1.276 | 3/14 | 0.21 | | TWN | 1960 | 0.002 | 0.768 | 506.50 | +1.216 | 1/14 | 0.07 | | SGP | 1965 | 0.026 | 0.850 | 33.26 | +1.174 | 3/14 | 0.21 | | CHN | 1978 | 0.018 | 0.549 | 30.85 | +0.688 | 4/14 | 0.29 |

Average ATT at 40-yr horizon: +1.088 log-points (~+197%). Cases ≥0.30 log-points: 4/4. Mean per-case placebo p: 0.20.

Per-case donor weights

KOR (treated 1961)

| Donor | Weight | |---|---:| | EGY | 0.919 | | PAK | 0.060 | | BRA | 0.021 |

TWN (treated 1960)

| Donor | Weight | |---|---:| | EGY | 0.552 | | BRA | 0.234 | | MEX | 0.083 | | TUR | 0.082 | | IDN | 0.049 |

SGP (treated 1965)

| Donor | Weight | |---|---:| | CHL | 0.447 | | THA | 0.399 | | BRA | 0.154 |

CHN (treated 1978)

| Donor | Weight | |---|---:| | IND | 0.746 | | GHA | 0.116 | | PAK | 0.072 | | IDN | 0.065 |

Method downgrade note

The YAML's synth_did template (Arkhangelsky-Athey-Hirshberg synthetic DiD, with both unit and time weights jointly optimised) is approximated here by per-case synthetic control averaged across cases. Inference uses each case's rank within its own donor-as-placebo distribution rather than the joint synth-DiD bootstrap. This is a power-losing simplification — disclosed.

The pre-registered Polity-attenuation gate (rerun on polity-positive donor subset only) is DEFERRED: Polity5 vintage not in data/vintages/. When that fetcher lands, the v1.1 rerun resolves the third falsification component.

Steelman-live concerns

  1. KOR/TWN/SGP/CHN starting incomes were already converging-leaders in 1960; selection on initial trajectory (Maddison 1950s data is sparse for several donors) may inflate ATT.
  2. Donor pool conflates Latin-American populist-mercantilist with South-Asian import-substituting and African post-colonial — heterogeneous controls.
  3. CHN 1978 is post-Cultural-Revolution rebound; pre-trend rebound dynamics inflate the gap independent of industrial policy content.
  4. Maddison 2020 ends 2018; modern East Asian convergence in the 2018-2024 window is not in this run.
  5. The hypothesis is in tension with trade_liberalisation_growth_effect; both stories can hold (selective-protection + openness combo).

Provenance

Reproduces from vintages in manifest.yaml. See replication.py.

Strongest opposing argument

Every hypothesis ships with its charitable opposing argument. The framework earns credibility by handling objections at their strongest, not weakest.

Notes

This hypothesis and trade_liberalisation_growth_effect are designed to tension against each other. The developmentalist story argues selective protection + export discipline matters; the liberalisation story argues openness matters. Both can hold simultaneously (selective protection for infant industry + openness elsewhere) but the v1 specs test the marginal claim of each. Results should be read jointly, not as either-or. Taiwan (TWN) data requires Maddison + IMF rather than WDI for the full span because WDI excludes Taiwan from standard tabulations.

Authored framework. Read the transparency note.