FINC3017 · Investments And Portfolio Management
Practical & Current Issues: Alternatives & Machine Learning
Practical & Current Issues: Alternatives & Machine Learning (Week 12) closes the course with two frontier topics. Alternative assets — hedge funds, private equity, real estate, commodities, infrastructure and crypto — offer low correlation and diversification but bring illiquidity, higher fees and valuation complexity, and private equity follows a recognisable life-cycle from angel through venture to buyout. Machine learning in finance splits into supervised methods (which predict a labelled outcome) and unsupervised methods (which find structure), and you manage the trade-off between Type-I and Type-II errors using an F-score that weights false positives against false negatives.
What this chapter covers
- 01Alternative assets: hedge funds, private equity, real estate/REITs, commodities, infrastructure, crypto
- 02Alts characteristics: illiquidity, low correlation, higher fees, complexity
- 03Private-equity life-cycle: angel → venture → buyout → growth/mezzanine
- 04Supervised ML (logistic regression, ridge, decision trees, SVM) vs unsupervised ML (clustering, PCA)
- 05The confusion matrix and Type-I (false positive) vs Type-II (false negative) errors
- 06F1 score = TP/(TP + ½FN + ½FP) and the F_β generalisation
- 07Weighting errors by investment risk via F_β
- 08Crash prediction via rolling logistic regression and dynamic de-risking
Reading a confusion matrix: precision, recall and F1
- 2 marksPrecision = TP/(TP + FP) = 30/(30 + 10) = 30/40 = 0.75 — of the crashes the model flagged, 75% were real.
- 2 marksRecall = TP/(TP + FN) = 30/(30 + 20) = 30/50 = 0.60 — of the real crashes, the model caught 60%.
- 2 marksF1 (balanced) = TP/(TP + ½FN + ½FP) = 30/(30 + 10 + 5) = 30/45 = 0.6667.
- 1 markCross-check via the harmonic mean: F1 = 2·P·R/(P + R) = 2 × 0.75 × 0.60/(0.75 + 0.60) = 0.90/1.35 = 0.6667 — the same value.
Key terms
- Alternative assets
- Investments outside traditional stocks and bonds — hedge funds, private equity, real estate, commodities, infrastructure and crypto. They offer diversification through low correlation but typically carry illiquidity, higher and more complex fee structures, and harder-to-value (often stale) prices.
- Private-equity life-cycle
- The financing stages a company passes through as a private-equity target: angel and seed funding, venture capital, leveraged buyout, and later growth or mezzanine financing. Each stage carries different risk, return and time-horizon characteristics for the investor.
- Supervised vs unsupervised learning
- Supervised methods (logistic regression, ridge, decision trees, SVM) learn to predict a labelled outcome from inputs; unsupervised methods (clustering, principal component analysis) find structure in unlabelled data. The split mirrors prediction versus exploration in finance applications.
- Type-I vs Type-II error
- A Type-I error is a false positive (flagging something that is not there); a Type-II error is a false negative (missing something that is). In a crash predictor a Type-II error — missing a real crash — is usually the more expensive mistake to make.
- F1 / F-beta score
- A single accuracy measure combining precision and recall: F1 = TP/(TP + ½FN + ½FP), the harmonic mean that weights the two errors equally. The F_β generalisation tilts the weighting toward recall (β > 1) or precision (β < 1) to reflect the relative cost of each error.
Practical & Current Issues: Alternatives & Machine Learning FAQ
Why do investors add alternatives despite their high fees and illiquidity?
The main draw is diversification: many alternatives have low correlation with public stocks and bonds, so adding them can lower overall portfolio risk and shift the efficient frontier outward. Investors accept the illiquidity, higher fees and valuation complexity as the price for that diversification and for access to return sources unavailable in public markets — but the exam expects you to weigh both sides.
What is the difference between supervised and unsupervised machine learning in finance?
Supervised learning uses labelled data to predict a known target — for example, logistic regression estimating the probability of a market crash. Unsupervised learning has no labels and instead finds structure, such as clustering stocks into groups or using PCA to extract common factors. Supervised is about prediction; unsupervised is about discovering patterns.
How does the Type-I / Type-II error trade-off shape a crash-prediction model?
A model can be tuned to catch more crashes (raising recall, fewer Type-II errors) at the cost of more false alarms (lower precision, more Type-I errors), or vice versa. Because missing a real crash is usually far more damaging than a false alarm, you bias the model toward recall — using an F_β score with β > 1 — and accept some extra false positives to avoid the costly misses.
Exam move
Be able to build precision, recall and the F1 score from a confusion matrix, and connect false positives and false negatives back to Type-I and Type-II errors from Week 1. On the conceptual side, know the trade-offs of alternative assets, the private-equity life-cycle, and the supervised-versus-unsupervised split, since this final week supplies quick conceptual MCQs that round out the closed-book exam.
