Appendix d — Data Card — International Macro: ECB SDW · BIS · IMF IFS · World Bank WDI · OECD.Stat

Provider & what it is. Five official statistical agencies that publish cross-country macroeconomic and financial series — the data you reach for when your question spans more than one country (exchange rates, interest rates, GDP, inflation, credit, debt, banking). ECB SDW (European Central Bank Statistical Data Warehouse, now the ECB Data Portal) serves euro-area and EU monetary, financial, and banking series. BIS (Bank for International Settlements) is the central banks’ bank: policy rates, effective exchange rates, credit-to-GDP gaps, debt securities, and the property-price and global-liquidity statistics. IMF IFS (International Financial Statistics, via the IMF Data portal) is the classic cross-country panel of balance-of-payments, monetary, exchange-rate, and national-accounts series. World Bank WDI (World Development Indicators) is the broad development panel — GDP, population, financial-sector depth, poverty, governance — for nearly every country and decades of years. OECD.Stat (now the OECD Data Explorer) covers OECD and partner economies across national accounts, labor, finance, and more. The reveal: these are the free public counterparts to Bloomberg’s international terminal — each is authoritative for its niche, none is a perfect superset of the others, and the entire craft is knowing which provider owns the series you need and joining them on consistent country codes.

Coverage. Country × indicator × time (mostly annual or quarterly; some monthly/daily for financial series). World Bank WDI: ~200+ countries, 1960–present, thousands of indicators. IMF IFS: ~190 members, monthly/quarterly/annual back decades. BIS: dozens of reporting countries, varying start dates by dataset. ECB: euro area / EU, with deep monthly financial detail. OECD: ~38 members plus partners. All are revised over time — pin a vintage where the level matters.

Key identifiers. The join key is the country code, and the standards differ by provider, which is the classic merge bug. World Bank uses ISO 3-letter codes (USA, DEU, JPN) and its own region/income aggregates; IMF uses ISO codes and its own numeric IFS country codes; OECD and ECB use ISO/2- or 3-letter plus SDMX dimension codes; BIS uses ISO 2-letter in many datasets. Series themselves are identified by an indicator code (World Bank: NY.GDP.MKTP.CD; IMF: an SDMX series key) and a frequency (A/Q/M). Build an explicit country-code crosswalk before merging across providers, and keep codes as strings.

Access path. All free, mostly no key. Several speak the SDMX standard, and pandasdmx/sdmx1 can talk to ECB, OECD, IMF, and others; the World Bank has a simple REST API (and the wbgapi/world_bank_data packages):

import requests
import wbgapi as wb   # World Bank — no key

# World Bank WDI: GDP (current US$) for a few countries, tidy long format
gdp = wb.data.DataFrame("NY.GDP.MKTP.CD", ["USA", "DEU", "JPN"], time=range(2015, 2024))

# IMF IFS via the SDMX/JSON API (no key); series key is provider-specific
imf = requests.get("https://dataservices.imf.org/REST/SDMX_JSON.svc/"
                   "CompactData/IFS/Q.US.PMP_IX",  # [CHECK] exact IFS series key
                   timeout=60).json()

# ECB Data Portal (SDMX): e.g., euro-area MRO policy rate
ecb = requests.get("https://data-api.ecb.europa.eu/service/data/FM/"
                   "B.U2.EUR.4F.KR.MRR_FR.LEV",   # [CHECK] exact key
                   params={"format": "jsondata"}, timeout=60).json()

Most need no credential. Where a provider does issue an API key (some OECD/usage tiers), read it from the environment, never hard-code it (CONVENTIONS §5).

License & note. These are open data with attribution terms that vary by provider. World Bank WDI is released under CC BY 4.0 — redistributable with attribution. OECD, ECB, IMF, and BIS publish under their own open-terms (generally free for non-commercial/research use with citation; some BIS series carry redistribution conditions) [CHECK exact license per provider before redistributing]. In practice for student research you may cache and share, but cite the originating provider and the access/vintage date — and where a provider re-publishes another’s series, cite the original (the same discipline FRED forces, per CONVENTIONS §6).

Gotchas. - Country codes do not match across providers. The number-one merge bug. DEU/DE/numeric IMF code all mean Germany; aggregates like “Euro area” or “World” sneak into country columns. Crosswalk explicitly; drop or label aggregates deliberately. - The same indicator is defined differently across sources. “GDP” can be nominal vs. real, local currency vs. USD vs. PPP, SA vs. NSA. Two providers’ “GDP growth” need not agree. Read the metadata, not just the column name. - Revisions and vintages. Macro series are revised; today’s value for 2020 is not what was published in 2021. For a real-time or forecasting design, you need the vintage, not the latest — the same look-ahead trap as FRED’s ALFRED (Ch 7.2). - Coverage holes and breaks. Poorer/smaller countries have gaps; definitional breaks (a euro changeover, a methodology revision, a country split) create discontinuities. A balanced-panel dropna() can silently bias your sample toward rich, well-measured countries. - SDMX keys are finicky. The dimension order and codes in a series key are provider-specific and easy to get wrong; verify against the provider’s data structure definition [CHECK exact keys above].

First 10 rows — schema sketch (illustrative; values invented, not real statistics).

iso3	country	provider	indicator_code	indicator	freq	period	value	unit
USA	United States	WorldBank	NY.GDP.MKTP.CD	GDP, current US$	A	2022	2.54e13	USD
DEU	Germany	WorldBank	NY.GDP.MKTP.CD	GDP, current US$	A	2022	4.08e12	USD
JPN	Japan	WorldBank	NY.GDP.MKTP.CD	GDP, current US$	A	2022	4.23e12	USD
USA	United States	IMF_IFS	FPOLM_PA	Policy rate	M	2023-11	5.33	percent
DEU	Germany	ECB	MRR_FR.LEV	ECB MRO rate	M	2023-11	4.50	percent
GBR	United Kingdom	BIS	CBPOL	Central bank policy rate	M	2023-11	5.25	percent
FRA	France	OECD	CPALTT01	CPI, all items	M	2023-11	4.0	pct YoY
JPN	Japan	IMF_IFS	ENDA_XDC_USD	Exchange rate, JPY/USD	M	2023-11	149.7	JPY/USD
BRA	Brazil	WorldBank	FR.INR.LEND	Lending rate	A	2022	39.6	percent
IND	India	WorldBank	NY.GDP.MKTP.CD	GDP, current US$	A	2022	3.39e12	USD

(Note the same iso3 join key spanning providers — and that the policy-rate definitions are not identical across IMF/ECB/BIS.)

Which chapter/lab/capstone uses it. These power any cross-country project. The methods are the panel and event-study tools from Week 3–4 (fixed effects, DiD) and the inference discipline of Week 5 (clustering — note these are country panels, so cluster thoughtfully, Ch 5.4 Petersen). A natural fit is a macro/monetary capstone in the spirit of Capstone 5 (FRED Macro Event Study) extended internationally — e.g., how do exchange rates or credit growth respond across countries to a policy shock — or any comparative finance question where one country is not enough. You acquire and document the series through the Ch 7.2 / Lab 7 reproducible-pull discipline: build the country-code crosswalk, pin the vintage, cache the raw responses, and cite each originating provider. (For U.S.-only macro, prefer FRED; reach for these when the question crosses borders.)

# Data Card — International Macro: ECB SDW · BIS · IMF IFS · World Bank WDI · OECD.Stat **Provider & what it is.** Five official statistical agencies that publish *cross-country* macroeconomic and financial series — the data you reach for when your question spans more than one country (exchange rates, interest rates, GDP, inflation, credit, debt, banking). **ECB SDW** (European Central Bank Statistical Data Warehouse, now the **ECB Data Portal**) serves euro-area and EU monetary, financial, and banking series. **BIS** (Bank for International Settlements) is the central banks' bank: policy rates, effective exchange rates, credit-to-GDP gaps, debt securities, and the property-price and global-liquidity statistics. **IMF IFS** (International Financial Statistics, via the IMF Data portal) is the classic cross-country panel of balance-of-payments, monetary, exchange-rate, and national-accounts series. **World Bank WDI** (World Development Indicators) is the broad development panel — GDP, population, financial-sector depth, poverty, governance — for nearly every country and decades of years. **OECD.Stat** (now the OECD Data Explorer) covers OECD and partner economies across national accounts, labor, finance, and more. The reveal: these are the *free* public counterparts to Bloomberg's international terminal — each is authoritative for its niche, none is a perfect superset of the others, and the entire craft is knowing which provider owns the series you need and joining them on consistent country codes. **Coverage.** Country × indicator × time (mostly annual or quarterly; some monthly/daily for financial series). World Bank WDI: ~200+ countries, 1960–present, thousands of indicators. IMF IFS: ~190 members, monthly/quarterly/annual back decades. BIS: dozens of reporting countries, varying start dates by dataset. ECB: euro area / EU, with deep monthly financial detail. OECD: ~38 members plus partners. All are revised over time — pin a vintage where the level matters. **Key identifiers.** The join key is the **country code**, and the *standards differ by provider*, which is the classic merge bug. World Bank uses **ISO 3-letter** codes (`USA`, `DEU`, `JPN`) and its own region/income aggregates; IMF uses ISO codes *and* its own numeric **IFS country codes**; OECD and ECB use ISO/2- or 3-letter plus SDMX dimension codes; BIS uses ISO 2-letter in many datasets. Series themselves are identified by an **indicator code** (World Bank: `NY.GDP.MKTP.CD`; IMF: an SDMX series key) and a **frequency** (`A`/`Q`/`M`). Build an explicit country-code crosswalk before merging across providers, and keep codes as strings. **Access path.** All free, mostly no key. Several speak the **SDMX** standard, and `pandasdmx`/`sdmx1` can talk to ECB, OECD, IMF, and others; the World Bank has a simple REST API (and the `wbgapi`/`world_bank_data` packages): ```python import requests import wbgapi as wb # World Bank — no key # World Bank WDI: GDP (current US$) for a few countries, tidy long format gdp = wb.data.DataFrame("NY.GDP.MKTP.CD", ["USA", "DEU", "JPN"], time=range(2015, 2024)) # IMF IFS via the SDMX/JSON API (no key); series key is provider-specific imf = requests.get("https://dataservices.imf.org/REST/SDMX_JSON.svc/" "CompactData/IFS/Q.US.PMP_IX", # [CHECK] exact IFS series key timeout=60).json() # ECB Data Portal (SDMX): e.g., euro-area MRO policy rate ecb = requests.get("https://data-api.ecb.europa.eu/service/data/FM/" "B.U2.EUR.4F.KR.MRR_FR.LEV", # [CHECK] exact key params={"format": "jsondata"}, timeout=60).json() ``` Most need no credential. Where a provider does issue an API key (some OECD/usage tiers), read it from the environment, never hard-code it (CONVENTIONS §5). **License & note.** These are *open data* with attribution terms that vary by provider. **World Bank WDI** is released under **CC BY 4.0** — redistributable with attribution. **OECD**, **ECB**, **IMF**, and **BIS** publish under their own open-terms (generally free for non-commercial/research use with citation; some BIS series carry redistribution conditions) `[CHECK exact license per provider before redistributing]`. In practice for student research you may cache and share, but **cite the originating provider and the access/vintage date** — and where a provider re-publishes another's series, cite the original (the same discipline FRED forces, per CONVENTIONS §6). **Gotchas.** - **Country codes do not match across providers.** The number-one merge bug. `DEU`/`DE`/numeric IMF code all mean Germany; aggregates like "Euro area" or "World" sneak into country columns. Crosswalk explicitly; drop or label aggregates deliberately. - **The same indicator is defined differently across sources.** "GDP" can be nominal vs. real, local currency vs. USD vs. PPP, SA vs. NSA. Two providers' "GDP growth" need not agree. Read the metadata, not just the column name. - **Revisions and vintages.** Macro series are revised; today's value for 2020 is not what was published in 2021. For a real-time or forecasting design, you need the vintage, not the latest — the same look-ahead trap as FRED's ALFRED (Ch 7.2). - **Coverage holes and breaks.** Poorer/smaller countries have gaps; definitional breaks (a euro changeover, a methodology revision, a country split) create discontinuities. A balanced-panel `dropna()` can silently bias your sample toward rich, well-measured countries. - **SDMX keys are finicky.** The dimension order and codes in a series key are provider-specific and easy to get wrong; verify against the provider's data structure definition `[CHECK exact keys above]`. **First 10 rows — schema sketch (illustrative; values invented, not real statistics).** | iso3 | country | provider | indicator_code | indicator | freq | period | value | unit | |---|---|---|---|---|---|---|---|---| | USA | United States | WorldBank | NY.GDP.MKTP.CD | GDP, current US$ | A | 2022 | 2.54e13 | USD | | DEU | Germany | WorldBank | NY.GDP.MKTP.CD | GDP, current US$ | A | 2022 | 4.08e12 | USD | | JPN | Japan | WorldBank | NY.GDP.MKTP.CD | GDP, current US$ | A | 2022 | 4.23e12 | USD | | USA | United States | IMF_IFS | FPOLM_PA | Policy rate | M | 2023-11 | 5.33 | percent | | DEU | Germany | ECB | MRR_FR.LEV | ECB MRO rate | M | 2023-11 | 4.50 | percent | | GBR | United Kingdom | BIS | CBPOL | Central bank policy rate | M | 2023-11 | 5.25 | percent | | FRA | France | OECD | CPALTT01 | CPI, all items | M | 2023-11 | 4.0 | pct YoY | | JPN | Japan | IMF_IFS | ENDA_XDC_USD | Exchange rate, JPY/USD | M | 2023-11 | 149.7 | JPY/USD | | BRA | Brazil | WorldBank | FR.INR.LEND | Lending rate | A | 2022 | 39.6 | percent | | IND | India | WorldBank | NY.GDP.MKTP.CD | GDP, current US$ | A | 2022 | 3.39e12 | USD | (Note the same `iso3` join key spanning providers — and that the policy-rate definitions are *not* identical across IMF/ECB/BIS.) **Which chapter/lab/capstone uses it.** These power any *cross-country* project. The methods are the panel and event-study tools from **Week 3–4** (fixed effects, DiD) and the inference discipline of **Week 5** (clustering — note these are country panels, so cluster thoughtfully, Ch 5.4 Petersen). A natural fit is a macro/monetary **capstone** in the spirit of **Capstone 5 (FRED Macro Event Study)** extended internationally — e.g., how do exchange rates or credit growth respond across countries to a policy shock — or any comparative finance question where one country is not enough. You acquire and document the series through the **Ch 7.2 / Lab 7** reproducible-pull discipline: build the country-code crosswalk, pin the vintage, cache the raw responses, and cite each originating provider. (For U.S.-only macro, prefer FRED; reach for these when the question crosses borders.)