Gains:
- Ability to systematically list the main limits of artificial intelligence in economic analysis (fabrication, data cut-off date, causality blindness, structural break)
- Ability to establish a secure usage and KVKK compliant workflow that protects confidential data, market-sensitive information and personal data
- Ability to design and implement an enterprise-level AI governance framework (registration, verification, accountability, approval chain)
Module Exam
1. A central banker puts an AI-generated inflation forecast directly into an official policy note without any verification. What is the fundamental mistake in this approach?
- A) Transforming the artificial intelligence output into a decision without connecting it to the source, recalculating it and undergoing economic verification; Ignoring that the responsibility lies with people ✔
- B) It is strictly forbidden to use artificial intelligence in economic forecasts
- C) Artificial intelligence always shows inflation lower than it is
- D) The prediction is not presented in a graph
Description: Artificial intelligence produces a statistical outline; but the responsibility for forecasting and the content of a formal policy brief rests with the competent economist and the institution. Unverified output is as risky as an unsigned report; The number must be linked to the original source, recalculated and passed through an economic filter.
2. What is the safest approach when you ask artificial intelligence, 'Give me the current account deficit figure for Türkiye in 2024'?
- A) Trust the number given by artificial intelligence and use it directly
- B) Confirming the figure from the CBRT's official balance of payments publication and verifying the date/unit ✔
- C) Getting the number from a random news site on the internet
- D) Ignoring the number because it will change anyway
Explanation: The number given by artificial intelligence may be older than the cut-off date of the training data or may be made up (hallucination). The only accurate source of economic data is the official publication; In this example, it is the CBRT balance of payments statistics. The number must be confirmed from the original source.
3. An analyst gives raw CPI series to artificial intelligence and says 'calculate real growth'. Which of the following is a correct validation behavior?
- A) Adding the result to the report without reading it
- B) Just looking at whether the outcome 'seems reasonable'
- C) Verify the result by reading the generated formula/code and manually calculating at least one period ✔
- D) Ask artificial intelligence 'are you sure?' and being satisfied with the answer 'yes'
Explanation: In standard economic calculations such as real-nominal conversion, the AI may not use the correct formula or mix up the base year. The analyst must read the generated code and formula, manually calculate at least one observation and compare the result. Transparent, explainable accounting is preferred.
4. Why is 'backtesting' done in time series forecasting?
- A) To show how 'nice' the model fits historical data
- B) To automatically finalize the estimate
- C) To beautify the data
- D) To measure the real prediction success by testing the model in previous periods when it was not in training ✔
Description: Backtest measures how well the model predicts over periods in the past that it does not know; It protects the model from the trap of overfitting by only looking at the data it fits. It is a prerequisite for the usability of the model in the real world.
5. What is the 'base effect' in inflation comments and why does it require attention?
- A) Annual change is affected by the low/high of last year's comparison period ✔
- B) Inflation always rises
- C) Core inflation being higher than the headline
- D) Seasonal adjustment of prices
Explanation: The base effect is that the annual rate of change is affected by last year's (base) low or high as well as this year's development. A decline in headline annual inflation may be due to last year's high base and not a real slowdown. This pitfall is often overlooked in AI summaries.
6. How can the role of artificial intelligence be most accurately defined in a scenario study?
- A) A seer who knows the future with certainty
- B) The final decision maker who determines the assumptions on his own
- C) An assistant that turns obvious assumptions into transparent calculations and makes options visible; judgment is in man ✔
- D) A tool of precision that makes the script unnecessary
Explanation: The scenario is not a prediction, but a conditional 'if-then' exercise based on explicit assumptions. AI is an accelerator in translating assumptions into spreadsheets and making options visible; However, ownership of the assumptions, their reasonableness, and final judgment rest with the economist.
7. AI uses strong causal language such as 'X increases Y' when interpreting a regression output. What is the correct economist response?
- A) Accepting the causal language as it is
- B) Correct the language by checking the risks of omitted variables/endogeneity and correcting the fact that relationship does not mean causality.
- C) Throwing away regression completely
- D) Assuming causality is proven if R-squared is high
Explanation: Regression mostly measures relationship (correlation); Causality can only be claimed under appropriate design (experiment, instrumental variable, natural experiment, etc.) and assumptions. There are risks of omitted variables, endogeneity, and reverse causality. The economist must curb this excessive causal language and question the conditions.
8. You had artificial intelligence prepare a literature summary on an economics topic, and the text gives references such as '(Yılmaz, 2019)'. What should you do first?
- A) Putting the citations in the report as they are
- B) Assuming good quality just because there are many citations
- C) Just looking at the consistency of publication years
- D) Verify from the original source that each reference actually exists and contains the finding ✔
Explanation: Artificial intelligence can produce sources and references that do not actually exist (hallucination). Each reference must be verified from the original source that it actually exists, that its attribution is correct, and that it actually contains the claimed finding. Unverified attribution is an academic and professional risk.
9. What is the most honest way to convey uncertainty in a public economic report?
- A) Giving a single precise number and not mentioning uncertainty at all
- B) Hiding the worst case scenario and giving only the optimistic number
- C) Clearly state the forecast range, scenarios and underlying assumptions ✔
- D) Dismissing uncertainty by saying 'it's complicated'
Explanation: Uncertainty is inevitable in economic forecasting and analysis. Clearly stating ranges, scenarios, and assumptions rather than a single exact number is honest communication. Hiding ambiguity misleads the reader and undermines the credibility of the organization.
10. When a chart shows GDP growth, it starts the vertical axis at 2% rather than 0%. Why might this be misleading?
- A) Dashed axis may exaggerate small differences and make the change larger than it is ✔
- B) Where the axis starts does not affect the graph at all
- C) The graph automatically becomes more accurate
- D) Only color selection is important, axis is unimportant
Explanation: A dashed axis (not starting from zero) can visually magnify small differences and make the change appear dramatic. In honest visualization, the choice of axis should not distort the message; If a dashed axis is used, it must be clearly marked.
11. A think tank analyst plugs market-sensitive data that has not yet been made public into a public (non-enterprise) AI tool. What is the key risk?
- A) There is no risk, the data is already numerical
- B) The only risk is that the graph turns out to be wrong
- C) Artificial intelligence processes data very slowly
- D) Risk of data leakage, processing by third parties and violation of privacy/market rules ✔
Disclosure: Entering confidential or market-sensitive data into a public tool creates the risk of the data being processed by third parties, leaked or used to train the model; This poses both a privacy/KVKK violation and a risk of market manipulation and insider information. Sensitive data should be processed in institution-approved, secure tools and anonymized when necessary.
12. The figures of the same indicator taken from two different sources (TURKSTAT and an international organization) do not match. What is the most likely innocent explanation?
- A) One of the sources is definitely lying
- B) There may be a difference in definition, base year, scope, seasonal adjustment or revision; metadata must be aligned ✔
- C) The difference in numbers is unimportant, taking their average is sufficient
- D) It is enough to ask the artificial intelligence and let it choose which one is correct.
Explanation: The difference in the same indicator in different sources is mostly due to differences in definition, base year, scope, seasonal adjustment or revision (version). Metadata (definition, unit, period, correction) must be aligned before comparing figures.
13. Which of the following is a task where AI is STRONG in economic analysis, accelerating with human validation?
- A) To give final approval of the official inflation figure alone
- B) Making monetary policy decisions instead of the institution
- C) Speed up tasks such as drafting data cleaning code and summarizing long reports ✔
- D) Putting the unverified guess in the public report
Description: Artificial intelligence; Strong in mechanical and drafting tasks such as drafting code, summarizing long report, suggesting data cleaning steps, correcting formatting and language. However, the ultimate responsibility for forecasting, policy decision and production of official figures rests with humans. The correct use is to accelerate where it is strong and keep the critical judgment in the person.
14. What is the most appropriate governance practice to ensure that artificial intelligence outputs are monitored and audited in an institution?
- A) Establish a traceability and approval chain that records the request, tool, data and verification steps ✔
- B) Not recording the outputs at all so that the workload does not increase
- C) Everyone can freely use the vehicle they want with the data they want.
- D) Only the most senior person uses artificial intelligence, no records are kept
Description: Sound governance; It requires recording which output was produced with which request, with which tool, with which data, a verification step and a responsible approval chain. This discipline of recording and acknowledging allows tracing errors back and clarifying responsibility.