Development Ladder

Method

The atlas build uses one time-consistent BACI HS92 V202601 classification for 1995-2024. Raw BACI values are thousands of USD; the country-frontier output reports world market size in millions of USD.

Exports are represented by three-year centered means. A product is present when Balassa RCA >= 1, absent and at risk of entry when RCA < 0.5, and in the buffer state otherwise.

The panel keeps countries with 2020 population >= 1 million, 2024 exports >= 1 billion USD and reports in all 30 years. Products must be present in at least 20 of 30 years.

Each transition window runs from t to t+5. Starts span 1996-2018 and are grouped into E1 1996-2003, E2 2004-2010 and E3 2011-2018.

Density is the Hidalgo et al. (2007) proximity-weighted share of related products already in the basket, with a separate proximity matrix estimated within each era. Density deciles use pooled all-era cutoffs.

Income groups are GDP-per-capita quartiles among panel countries at each window start, not World Bank income groups.

The directed HS4 atlas estimates P(enter j within five years | exporting i, not exporting j) against the base entry rate into j. Published cells require n_at_risk >= 30 and n_entered >= 8 and are ranked by the Wilson lower-bound lift.

The country strategy brief deliberately recomputes a current HS4 Balassa basket from country_year_product at BACI_YEAR_MAX. It aggregates exports to HS4 before calculating RCA and does not read the HS6 rca_matrix.

Equations

Validation

• The ledger reports a verified decline in five-year entry probability from positioned baskets: 5.33% for E1, 4.47% for E2 and 3.69% for E3.
• Pipeline QA and an independent check reproduced the VNM 8517 entry and BGD frontier market sizes.
• The current-basket correction was independently recomputed: BGD has 87 HS4 headings with RCA >= 1 and IND has 437.
• The shipped outputs contain three summary-era rows, 150 curve cells, 72,710 directed HS4 edges, 120,849 country-path rows and 6,700 country-frontier rows.
• The curve output stores Wilson lower and upper bounds alongside n_at_risk and n_entered.

Limitations

• Entries are RCA crossings, not plant openings. RCA can move because its country or world denominators move.
• The estimates are descriptive conversion rates, not causal effects.
• Era proximity matrices differ by construction. Pooled density cutoffs make positions comparable, but the underlying proximity relationships remain era-specific.
• Income quartiles are relative positions within the panel and should not be read as World Bank income groups.
• The selected panel excludes smaller, lower-export and incomplete-reporting economies.
• n_at_risk and n_entered count overlapping country-year windows, not distinct countries. The overlap makes interval precision optimistic, and the last complete entry window closes in 2018.
• The country brief uses raw latest-year HS4 RCA, while the atlas itself uses three-year mean states at 2023. The current basket and historical atlas are therefore not constructed identically.
• Point lifts are selected from the strongest available edges and can carry winner's-curse bias.
• The Method text on /ladder says directed cells require three entries, but the ledger and the page's empty-state copy say eight. The implementation-backed card uses eight.

References

• Balassa (1965), Trade Liberalisation and Revealed Comparative Advantage.
• Hidalgo, Klinger, Barabasi and Hausmann (2007), The Product Space Conditions the Development of Nations.
• CEPII BACI HS92 V202601.
• World Bank World Development Indicators.

Endpoints