Portfolio optimisation through machine learning - a structured programme for working professionals
Quickprem brings together analysts, quants, and portfolio managers to study how modern ML methods change the way portfolios are built, stress-tested, and rebalanced under real market conditions.
Grounded in current practice
The curriculum reflects what practitioners are actually doing now
The material is updated each cohort based on what participants bring from their own desks - live questions from active portfolio work, not textbook scenarios. Recent modules have covered regime detection using hidden Markov models, factor exposure management with LASSO regularisation, and covariance estimation under non-stationarity.
Instructors hold active roles in asset management and quantitative research. They reference specific tools - scikit-learn, PyPortfolioOpt, cvxpy - and discuss failure modes they have seen in production environments, not just theoretical ones.
This programme is specific about who it is built for
There is no universal fit. The content assumes a certain baseline and moves quickly past it. Knowing whether you belong here before you apply saves time for everyone.
- Work with financial data daily - equity, fixed income, or multi-asset
- Write Python with some comfort - you do not need to be a software engineer
- Have hit the limits of mean-variance optimisation and want to go further
- Can commit 6–8 hours per week for the duration
- Want to apply what you learn to a real portfolio problem, not a toy dataset
- Are new to finance and looking for an introduction to investing
- Have no prior exposure to statistics or linear algebra
- Expect a tool that generates trading signals automatically
- Cannot dedicate consistent time during the eight-week period
What moving through this actually feels like
Weeks 1–2
Orientation through friction
The first two weeks are deliberately uncomfortable. You encounter problems from your own field reframed through an ML lens - and the gap between what you know and what the model requires becomes visible. That gap is the starting point.
Weeks 3–5
Building with real data
You move from theory into implementation. Sessions focus on specific decision points - how to handle look-ahead bias, when regularisation helps and when it masks the problem, how to validate a model that cannot be backtested cleanly. Peers review your work and push back on assumptions.
Weeks 6–7
Stress-testing your approach
Instructors present edge cases drawn from actual portfolio failures. You test your models against them. The goal is not to break your work - it is to understand where the boundaries of your method are before you apply it in a live context.
Week 8
Applied project and practitioner review
You present a project built around a portfolio problem you chose in week one. Reviewers are practitioners, not academics. Their feedback addresses whether the approach would hold in a real investment process - and what would need to change if it would not.
The people and context around you during the programme
Learning alongside practitioners from similar roles changes what questions get asked. Cohorts draw from asset managers, pension funds, and quantitative research teams across southern Africa - people dealing with the same data constraints and regulatory environment you work in.
Cohort of peers, not strangers
Each intake is capped so that discussion stays substantive. You will know who else is in the room and what they work on. Most cohort relationships extend beyond the eight weeks.
Facilitated discussion, not lectures
Sessions are structured around problems, not slides. Instructors prompt, challenge, and redirect. You are expected to contribute, not just absorb - which is uncomfortable and useful.
Regional professional context
The Johannesburg financial market has specific characteristics - liquidity constraints, currency exposure, JSE data availability. Instructors address these directly rather than defaulting to US or European examples.
What stays with you after the programme closes
Skills that only apply inside a classroom are not worth much. The table below maps what participants carry forward against where those capabilities show up in practice.
| Capability built | Applies in live portfolio work | Transferable to new roles | Deepened by continued practice |
|---|---|---|---|
| ML-based covariance estimation | |||
| Regime detection and signal conditioning | |||
| Constraint-aware optimisation with cvxpy | |||
| Model validation under non-stationarity | |||
| Peer critique and assumption auditing | |||
| Practitioner network access |
Stated in their own words, not summarised
"I came in expecting to learn a few new techniques. What actually happened was that I started seeing the assumptions underneath the tools I had been using for years - and realising which ones were doing quiet damage to my risk estimates. That was uncomfortable and genuinely useful."
Adaeze Nwofor
Senior Portfolio Analyst, fixed income, Johannesburg
"The peer review sessions were the part I did not expect to value. Having someone with a different background challenge your model choices forces you to actually defend them - or admit they were just defaults."
Liesl Badenhorst
Quantitative researcher, multi-asset, Cape Town
"I had read about LASSO regularisation but never understood when not to use it. The session on sparse models in low-liquidity environments finally made that distinction concrete."
Tendai Mhuriro
Portfolio manager, equity, Pretoria