Where portfolio decisions get sharper.
Quickprem runs structured online seminars on machine learning applied to portfolio optimisation - for practitioners who want to understand the mechanics, not just the outputs.
started operating
Programme depth
Numbers that describe how we structure learning
Each figure below reflects a deliberate choice about how seminars are built - depth over breadth, specificity over generality.
Core seminar modules
Each module addresses a distinct problem in ML-based portfolio construction - from feature engineering to out-of-sample validation.
Live discussion sessions per track
Structured peer exchanges where participants work through the same dataset using different model assumptions and compare results.
Programme tracks available
Foundations, applied ML allocation, and quantitative risk management - each track builds on the previous without repeating material.
Years of active delivery
The curriculum has been revised through real participant feedback across multiple cohorts - not a static product built once and left.
What Quickprem actually teaches - and what it does not
Machinelearninginportfoliomanagementisatechnicaldiscipline,notashortcut.
The seminars focus on the gap between knowing that a method exists and being able to apply it to real allocation problems. Topics like regularised covariance estimation, hierarchical risk parity, and regime-aware rebalancing are covered with working code examples and annotated case studies - not slide decks of definitions.
Participants are expected to bring familiarity with basic statistics and Python. The sessions do not teach programming from scratch. What they do is show how specific ML tools interact with portfolio constraints, transaction costs, and the structural properties of financial return data.
Covariance estimation under non-stationarity - Ledoit-Wolf shrinkage, graphical lasso, and when each applies to equity portfolios.
Supervised learning for factor timing - gradient boosting and LSTM approaches tested against naive equal-weight benchmarks.
Backtesting discipline - walk-forward validation, purged cross-validation, and how to avoid look-ahead bias in time-series data.
Drawdown-aware objective functions - replacing variance with CVaR or maximum drawdown in the optimisation target.
Pre-seminar prep
Reading list and a dataset distributed one week before. Participants arrive with context, not cold.
Live analysis sessions
Facilitator-led walkthroughs with open Q&A. Specific model choices are challenged, not just presented.
Peer review exercise
Small groups critique each other's portfolio construction decisions using a structured rubric.
