FIN306 in one sentence

Data science is statistical science: the disciplined study of variation and uncertainty in financial data.

Three challenges from Week 1, applied every week after (Gelman, Hill, and Vehtari 2020):

• Sample to Population: do my data represent the market I claim?
• Treatment to Control: would the result hold without the signal?
• Measurement to Construct: am I measuring what I think I am?

Bias, variance, and the virtue of complexity

• Bias-variance trade-off (Week 1): no model is free; we choose where to pay
• More parameters can help out-of-sample if we regularise (Kelly, Malamud, and Zhou 2024)
• This reframes “more data” and “more features” from danger into opportunity
• Direct implication for Scaffold B: tree ensembles vs OLS on JKP UK (Gu, Kelly, and Xiu 2020)

Why FinTech exists: the cost puzzle

• The unit cost of US financial intermediation has been roughly constant for 130 years (Philippon 2016)
• Technology has reshaped retail, telecoms, and manufacturing; finance has resisted
• That stubborn cost is the opportunity FinTech firms chase
• It is also the first place to be sceptical: easy gains have been claimed many times before

Data quality: where models go wrong before they start (Week 2)

• Survivorship bias: today’s index hides yesterday’s failures
• Look-ahead bias: information you would not have had in real time
• Selection bias: who shows up in your data, and who is missing
• Measurement to construct gap: VIX is implied volatility of an option set, not “fear”

Your CW1 risk register was the first formal practice of this discipline. CW2 Section B should sound the same.

Predicting return vs predicting risk (Weeks 3 and 4)

• Returns (ARIMA, Week 3): tiny R², near-zero predictability, and that is honest
• Risk (GARCH, Week 4): meaningful R², because volatility clusters
• The lesson is not “models fail”; it is what they can and cannot do
• VIX example: a good measurement, a contested construct

Robo-advisors and portfolio reality (Week 5)

• Mean-variance optimisation (Markowitz 1952) is fragile under estimation error
• Robo-advisors widen access to portfolio construction (Reher and Sokolinski 2024)
• Live questions: governance, suitability, fee transparency
• The Bloomberg lab made the estimation problem concrete

Alternative finance and credit scoring (Week 6)

• Marketplace lending data is selection-biased by design (who applies, who is funded)
• Digital footprint variables (device type, email provider) add about 1.3pp AUC over standard credit scores (Berg et al. 2020)
• That is real, and also opaque to the borrower
• Inclusion and fairness are both genuine concerns; they are not the same concern

Cryptocurrency and rare-event detection (Week 8)

• Crypto microstructure: thin books, regime shifts, fat tails
• Fraud detection as rare-event classification: imbalanced classes, walk-forward CV, cost-sensitive thresholds
• Class imbalance changes which metric matters: AUC, precision-recall, expected cost
• Direct anchor for Scaffold A (Elliptic Bitcoin, around 10 percent illicit)

Factor investing and the replication crisis (Week 9)

• Fama-French is a starting point, not an end state
• Harvey, Liu and Zhu: the factor zoo; the right t-hurdle is above 3 (Harvey, Liu, and Zhu 2016)
• Jensen, Kelly and Pedersen: a careful global re-test (Jensen, Kelly, and Pedersen 2024)
• Direct anchor for Scaffold B (JKP UK monthly factors with tree models and SHAP)

Backtesting, validation, sequence learning (Weeks 10 and 11)

• Walk-forward CV with embargoes; in-sample is not evidence
• PBO from CSCV (Bailey et al. 2015); DSR (Bailey and Prado 2014): the gap measures the cost of search
• Sequence learning showed why temporal structure matters in finance
• For Scaffold C (volatility), evaluate forecasts with Mincer-Zarnowitz, not Sharpe

Algorithmic accountability in lending

• Digital footprint scoring (Berg et al. 2020) is opaque to the borrower by construction
• Mortgage technology can be predictably unequal across protected groups (Fuster et al. 2022)
• The accountability question: when a model denies credit, who is on the hook?
  • Data scientist? Product owner? Vendor? Risk function? Regulator?
• “Compliance with the rules” and “treats customers fairly” are not the same test

Transparency vs interpretability

• SHAP (used in Scaffold B) gives feature attributions; it does not give causal explanations
• Quoting model output is not the same as explaining a decision
• The CW2 report tests this skill: tell the marker what your model cannot justify
• Fairness has multiple, mathematically incompatible definitions; the choice is normative

Regulating AI in finance: the UK lens

• EU AI Act (Regulation 2024/1689): credit scoring and market surveillance fall under high-risk obligations
• UK FCA: principles-based; supervisory sandboxes; consumer duty
• Bank of England SS1/23: model risk management principles for PRA-regulated firms
• “AI vs no AI” is the wrong question; “governable AI vs ungoverned AI” is the real one

The scarce skill: judgement under uncertainty

• LLMs can produce output; they cannot be accountable for it
• FIN306 trains the rarer skill: deciding what matters, weighing evidence, communicating limits
• That is also what employers in FCA-regulated firms hire for
• It is the same habit your CW1 risk register and CW2 limitations section reward

FIN306 schedule: this week and next

• Week 12 (today): Module synthesis; CW2 finalisation (see module schedule)
• Week 13: Coursework 2 due. Confirm date, time, and portal on Blackboard (module outline currently lists 8 May 2026, 23:59; always verify live)
• What you submit: one completed scaffold notebook plus a reflective report (2,500 words, +10% rule on the brief). Full file types and naming on Blackboard

FIN306 Coursework 2: Applied Data Science with Critical Reflection

• Weight: 70% of the module (CW1 was the recorded presentation and risk register)
• You chose one scaffold in Week 8; you complete the TODO sections and write the report. We assess judgement and reflection, not coding from scratch
• Scaffold A, Blockchain fraud (Elliptic): rare-event classification, walk-forward CV, cost-sensitive decisions; ties to fraud week
• Scaffold B, Tree-based factor investing (JKP UK): ensembles, SHAP, OOS evaluation; ties to factor week
• Scaffold C, Volatility forecasting: GARCH family, forecast evaluation; ties to volatility week and Bloomberg-style data

How FIN306 Coursework 2 is marked

Week 13 checklist (before you submit)

• Notebook: every TODO completed; Kernel → Restart & Run All with no errors
• Report: method rationale, data quality choices, results, limitations, responsible deployment. Use CW1 feedback where relevant
• Word count within brief (remember what counts toward the limit)
• Both files uploaded as Blackboard requires (e.g. .ipynb plus .docx; confirm)
• References complete; no placeholder citations
• Any AI-use or integrity statement required by the brief is included

Career Pathways in FinTech and Financial Data Science

Roles:

1. Data Science / Quantitative Analysis - Develop trading algorithms, risk models, fraud detection - Requires: Statistics, ML, programming, financial knowledge - Employers: Asset managers, banks, hedge funds, FinTech startups

2. Product Management - Design financial products/services, translate needs to specifications - Requires: Technical understanding, business acumen, communication - Employers: FinTech companies, banks’ digital divisions

3. Regulatory / Compliance - Navigate regulation, design compliance systems, engage regulators - Requires: Legal/regulatory knowledge, risk management, technology understanding - Employers: All financial institutions, consultancies, regulators

4. Research / Consulting - Analyse industry trends, advise firms/governments on strategy - Requires: Analytical skills, industry knowledge, communication - Employers: Central banks, think tanks, consultancies, academia

Final Reflection and Moving Forward

What we’ve covered (FIN306 arc):

• Foundations: statistical science, returns, bias–variance, honest inference
• Data quality and measurement (Bloomberg and structured labs)
• Time series and volatility modelling
• Robo-advisory and alternative finance
• Machine-learning bridge and CW2 scaffolds (Weeks 7–8)
• Cryptocurrency, fraud detection, factor investing, backtesting and validation
• Sequence learning (Week 11), then synthesis, ethics, and futures (this week)
• Ethical frameworks evaluating FinTech impacts

What remains uncertain:

• Will decentralisation succeed or will re-intermediation dominate?
• How will AI transform finance and employment?
• Will regulation enable innovation or stifle it?
• How will society navigate privacy vs surveillance?
• Will FinTech reduce inequality or increase it?

Your role:

You’ll shape answers through your work: building systems, analysing impacts, influencing policy, or leading organisations. This course provided frameworks, knowledge, and skills; now you apply them making financial services more efficient, accessible, fair, and stable.