Q1. Which of the following best describes the role of AI in finance?

1. Creating purely theoretical mathematical models with no data.
2. Using data-driven methods to support and automate financial decision-making, subject to risk and regulatory constraints.
3. Replacing all human employees in banks with robots.
4. Designing user interfaces for mobile banking apps only.

Select one:

A
B
C
D

Q2. Which of the following is not a typical property of financial time-series?

1. Non-stationarity.
2. Volatility clustering.
3. Perfectly Gaussian returns with no extreme values.
4. Regime shifts and structural breaks.

Select one:

A
B
C
D

Q3. Which combination best characterises panel data in credit risk modelling?

1. Single customer observed at a single time point.
2. Many customers observed over multiple time periods with repeated measurements.
3. Only macroeconomic variables observed monthly.
4. Only market prices at intraday frequency.

Select one:

A
B
C
D

Q4. (Short answer) Briefly explain what is meant by the signal-to-noise ratio in financial time-series, and why it matters for AI models.

Q5. In credit risk modelling, Expected Loss (EL) is usually defined as:

1. EL = PD + LGD + EAD
2. EL = PD × LGD × EAD
3. EL = PD / (LGD × EAD)
4. EL = LGD × EAD only, without PD

Select one:

A
B
C
D

Q6. Which model family is particularly popular and effective for tabular financial data such as credit risk features?

1. Gradient boosted decision trees (e.g. XGBoost, LightGBM).
2. Pure convolutional neural networks for images.
3. k-means clustering.
4. Fourier transforms only.

Select one:

A
B
C
D

Q7. Credit default and fraud events are typically:

1. Balanced classes with roughly 50% positive and 50% negative labels.
2. Highly imbalanced, with very few positive events relative to non-events.
3. Unlabelled and impossible to model.
4. Always removed from the dataset before modelling.

Select one:

A
B
C
D

Q8. (Short answer) Give one technique to address class imbalance in credit default or fraud detection models, and briefly describe how it helps.

Q9. Which of the following is a classical model specifically designed to model time-varying volatility in financial returns?

1. Logistic regression.
2. GARCH.
3. Naïve Bayes.
4. k-means clustering.

Select one:

A
B
C
D

Q10. In algorithmic trading, which statement best describes a mean-reversion strategy?

1. It assumes prices have no structure and move randomly.
2. It assumes prices that deviate from a long-term equilibrium tend to revert back.
3. It assumes all assets always trend in the same direction indefinitely.
4. It only trades based on macroeconomic news releases.

Select one:

A
B
C
D

Q11. Which of the following is a major risk when designing trading strategies with AI?

1. Underfitting in backtests but overperformance in live trading.
2. Backtest overfitting and data snooping leading to unrealistic performance expectations.
3. Having too few hyperparameters.
4. Excessive interpretability of the model.

Select one:

A
B
C
D

Q12. (Short answer) Briefly describe why transaction costs and slippage must be included when evaluating AI-driven trading strategies.

Q13. Which of the following is a typical example of financial text used in NLP models?

1. Earnings call transcripts.
2. Weather sensor readouts.
3. Raw audio waveforms only.
4. Satellite spectral bands.

Select one:

A
B
C
D

Q14. Which statement best captures the idea of alternative data in finance?

1. Alternative data refers only to historical stock prices.
2. Alternative data includes non-traditional sources such as web traffic, satellite imagery, credit card spending, and social media activity.
3. Alternative data consists exclusively of central bank announcements.
4. Alternative data is any dataset that has been manually typed by human analysts.

Select one:

A
B
C
D

Q15. Which of the following is a key risk associated with using Large Language Models (LLMs) for financial analysis?

1. They are unable to generate fluent text.
2. They may hallucinate plausible but factually incorrect information, which could mislead risk or investment decisions.
3. They can only work with structured numerical tables.
4. They are guaranteed to be unbiased by design.

Select one:

A
B
C
D

Q16. (Short answer) Give one example of how sentiment analysis on financial news or earnings calls can be integrated into a trading or risk model.

Q17. Which of the following is most closely associated with model risk management (MRM) in financial institutions?

1. Optimising user interface colours.
2. Governance and control processes that manage the risk that models are incorrect, misused, or poorly understood.
3. Only increasing model complexity without validation.
4. Delegating all decision-making to external vendors.

Select one:

A
B
C
D

Q18. Why is human-in-the-loop oversight important for AI systems in credit approvals or fraud detection?

1. Because regulators generally prohibit any use of data in decisions.
2. Because humans are always more accurate than models.
3. Because high-stakes decisions often require human review, the ability to override, and a check on model failures or bias.
4. Because AI models cannot operate 24/7.

Select one:

A
B
C
D

Q19. Which of the following is an example of a fairness or bias concern in AI for finance?

1. A model that uses only purely random numbers.
2. A credit scoring model that unintentionally produces higher decline rates for certain demographic groups, even when income and risk are similar.
3. A trading model that uses only intraday prices.
4. A volatility model that ignores macroeconomic data.

Select one:

A
B
C
D

Q20. (Short answer) Describe one way in which responsible AI practices can become a competitive advantage for a financial institution.