Units
1. Introduction to Artificial Intelligence in Economics: Roles, Boundaries, Validation and Data Ethics 2. Economic Data Collection and Source Verification: TURKSTAT, EVDS, IMF, World Bank, FRED 3. Data Cleansing, Transformation and Exploratory Analysis (With Python/pandas Support) 4. Time Series Analysis and Forecast Support 5. Interpreting Macroeconomic Indicators: Inflation, Growth, Unemployment, Current Account Balance 6. Scenario Modeling and Sensitivity Analysis 7. Econometric Model Support: Regression, Causation and Interpretation 8. Literature Review and Economic Research Synthesis 9. Policy Note, Report and Briefing Writing 10. Economic Data Visualization and Dashboard 11. Boundaries, Model Risk, Ethics, Privacy and Governance
Unit 2 / 11

Economic Data Collection and Source Verification: TURKSTAT, EVDS, IMF, World Bank, FRED

Gains:

  • Ability to compare major economic data sources (TURKSTAT, CBRT EVDS, IMF WEO, World Bank, Eurostat, FRED) according to purpose, scope, update frequency and reliability and choose the right source
  • Ability to use artificial intelligence for data source finding, serial code/description matching and download code generation, and confirm each result with the official source
  • Ability to see how version (revision), base year, unit and definition differences disrupt the analysis and gain the habit of verifying metadata

Every economic analysis starts with data, and the quality of the analysis can never exceed the quality of the data. The principle of "garbage in, garbage out" operates inexorably in economics: even the most elegant model, based on the wrong base, wrong definition, or wrong period, is wrong. In this unit, we will use artificial intelligence to find the right data, match serial codes and write the download code; But one golden rule will not change: The figure comes from the official primary source, not from artificial intelligence. AI can show you the way to the source; you get the data itself from that source.

Let's get to know the main sources

An economist's compass includes the following resources. The purpose, scope and update rhythm of each is different.

TURKSTAT (Turkish Statistical Institute). Türkiye's official statistical authority. Inflation (CPI/PPI), GDP, labor force, foreign trade, industrial production, population. This is the primary source for figures specific to Türkiye.

CBRT EVDS (Electronic Data Distribution System). Data portal of the Central Bank. Exchange rate, interest, money supply, balance of payments, current balance, reserves, expectation surveys. The first stop for financial and monetary data.

IMF (International Monetary Fund). Especially the WEO (World Economic Outlook) database: cross-country comparative GDP, inflation, debt, current account balance and projections. Ideal for international comparison; But projections are predictions, not realizations.

World Bank (World Bank — WDI). World Development Indicators: development, poverty, education, health, infrastructure indicators; long historical series and country comparisons.

Eurostat. Statistical office of the European Union; Harmonized (comparable) data for EU countries. Series such as HICP (harmonized inflation) are used in EU benchmarking.

FRED (Federal Reserve Economic Data). St. Louis Fed's massive series; Offers fast and programmable access (API) for US and global macro-financial data.

OECD. Comparative indicators, leading indicators, productivity and welfare series for developed economies.

Tip: The answer to the "correct source" question depends on the purpose. TURKSTAT for Türkiye's official inflation; IMF/World Bank to compare Türkiye with 30 countries; FRED for US interest. Taking the same indicator from the wrong source is a silent mistake.

Silent error sources: revision, base year, unit, description

The most dangerous errors in economic data are noiseless. Get to know four:

  • Revision (version). Many indicators (GDP, balance of payments) are revised months after they are first announced. The "provisional" figure today, the "final" figure after three months will be different. Know which vintage (version) you are using.
  • Base year. Indices are established based on a base year (=100). It is wrong to directly compare two series with different base years.
  • Unit. Million or billion, TL or dollar, current or fixed price? Unit confusion is the most common analysis error.
  • Definition/scope. What is the definition of "unemployment"? What items does "core inflation" exclude? Two series with the same name may measure different things.
Caution: If the figure for the same indicator from two sources does not match, your first assumption should not be "someone is lying" but "there is a difference in definition/base/scope/revision". Align the metadata (data citation) first.

Step by step: AI-powered data collection

  1. Clarify the need. Which indicator, which country/province, which period, which frequency (monthly/quarterly/annual), which definition?
  2. Select source. Explain the purpose to the AI ​​and suggest the appropriate primary source and full series name.
  3. Validate metadata. Confirm the unit, base year, description and update date of the selected series on the official page.
  4. Generate the download code. For those with an API, like FRED/World Bank, write Python code to the AI; Remove the download steps for TURKSTAT/EVDS.
  5. First check. After the data arrives, check the number of rows/columns, missing values, outliers and date range.
  6. Confirmation of source. Compare at least a few observations with the official publication; Does it match exactly?

three mini cases

Case 1 — Base year trap. One analyst obtained an industrial production index from two separate studies; one was based on 2015=100 and the other was based on 2021=100. “Combine these,” he told the artificial intelligence. If he skipped the metadata check, he would put the indices side by side and see a fake jump. He caught the base difference in the control step, turned the two into a common base and then combined them. The fake bounce is gone.

Case 2 — Fabricated API output. A researcher said to artificial intelligence, "Draw Türkiye's GDP per capita from the World Bank"; The AI ​​returned both the code and a table of numbers as "sample output". The researcher did not use the table; just ran the code and fetched the data from the real API. Then he realized that the numbers written by the artificial intelligence as "examples" were actually 10-15 percent off. Code tool, not number; This was the correct behavior.

Case 3 — Revision difference. In a note published six months ago, an institution wrote that the quarterly growth was 4.0 percent; In the current TÜİK data, the same quarter was revised to 3.5 percent. In the new report, the analyst stated in a footnote which version he used and explained the difference between the two figures with a revision. Transparency preserved credibility.

Source selection table

Purpose

Appropriate primary source

Attention

Türkiye inflation (CPI/PPI)

TURKSTAT

Base year, core/headline distinction

Exchange rate, interest, current balance, reserves

CBRT EVDS

Revision, series definition

GDP/inflation comparison between countries

IMF WEO

Projection ≠ realization

Development/long historical series

World Bank WDI

Coverage, missing years

EU harmonized indicators

Eurostat

HICP definition

US/global macrofinance, API

FRED

Seasonal adjustment status

Four copyable templates

1) Source and series matching:

I need the following indicator: [indicator, country/province, period, frequency, description]. Recommend the most appropriate primary official source and full series name for this. List the points I need to verify about the unit, base year, and description of the series. Do not give numbers; Just tell me where and how to buy it.

2) FRED/World Bank download code:

Write a code in Python with the [FRED/World Bank] API that downloads the following series: [series code/name], [date range]. The code takes the data into pandasDataFrame, prints the first/last 5 rows and the number of missing values. Don't give me an example/made-up figure; just give working code.

3) Metadata validation list:

I will use the following series: [series]. Create a checklist of metadata you need to verify before use: volume, base year, seasonal adjustment status, definition/scope, update date, revision status. For each item, write "what and where to confirm" in one sentence.

4) Two resource mismatch diagnoses:

I got the same indicator from two sources and the numbers are different: Source A: [value, imprint]. Source B: [value, imprint]. List possible innocent reasons for this difference (definition, base, scope, seasonal adjustment, revision, currency) and tell me how to check for each. Don't say "one is wrong"; First help me explain the discrepancy.

Weak prompt / Strong prompt

Weak prompt:

Give me Türkiye's growth data for the last 10 years.

The AI ​​pours out a picture from its memory, possibly old and partially fabricated; There is no source or imprint.

Powerful prompt:

I will use the 2014-2023 Türkiye annual real GDP growth. Tell me the primary source for this (TURKSTAT) and the full series name; tell me if the series is fixed price and where to look to verify the base year. Also give the steps/codes to download this data. Don't produce the numbers themselves.

Common mistakes

  • Mistaking the "sample output" table of artificial intelligence for real data. Those numbers may be fabricated; just use the code/guideline.
  • Comparing two series without aligning metadata. Difference in base/unit/definition produces spurious results.
  • Mistaking a projection for reality. Next year's figure in the IMF WEO is an estimate.
  • Ignoring the revision. Not stating which version you are using.
  • Frequency mismatch. Combining a monthly series with a quarterly series without correction.
  • Not citing the source in footnotes. A figure that is not traceable is a figure that is untenable.

In summary

The power of analysis is the power of data. Use AI to find the right source, match the series and write a download code; but always get the figure from the primary official source (TURKSTAT, EVDS, IMF, World Bank, Eurostat, FRED). Neutralize silent error sources such as revision, base year, volume, and description by aligning metadata. If two sources conflict, compare the attribution first.

Application task

Choose an indicator that you use frequently in your business. Have the AI ​​extract the source and metadata checklist with templates 1 and 3 above. Then find the same indicator from two different sources (e.g. TURKSTAT and World Bank) and compare the figures; If there is a difference, diagnose the cause with template 4 and note your finding in three sentences.

checklist

  • [ ] I chose the primary source suitable for my purpose (I made the distinction between Türkiye's official / international comparison).
  • [ ] I verified the unit, base year, description and update date of the series from the official page.
  • [ ] I used the data I took from the real source, not the "sample figures" given by artificial intelligence.
  • [ ] I compared at least a few observations verbatim with the official publication.
  • [ ] I checked the revision/version status and noted the version I was using.
  • [ ] I explained the two source conflicts by aligning the metadata.
  • [ ] I made the source of each figure I used traceable with a footnote.
Previous Next