I’m an economist specialising in AI machine learning applications for the fund management industry
I increase my clients’ productivity and profits by bringing the latest machine learning techniques into their regular business routines
I try hard to make that process painless, transparent, understandable and even fun. And I like to keep my clients in control of the journey, the method, the ethics and the impact on their organisation.
I started out analysing and forecasting market outcomes in industry (telecoms and air transport) before transferring these skills to consultancies and think-tanks. Along the way, I’ve had the privilege of working closely with global corporations, SMEs and start-ups.
My work is now in helping companies exploit their often-neglected goldmines of data and information to become more competitive and efficient. For me, the reward is in creating value by uncovering solutions that my clients don’t have the time or resources to find.
For fund managers this might mean forecasting option prices with neural networks, asset price forecasting through deep learning or tracing efficient portfolio frontiers using Monte Carlo simulation. For marketers, maybe predicting customer engagement likelihood with random forests or creating new customer segmentation by k-means.
The ground is always changing with machine learning. New software libraries and data sources are opening all the time. Algorithm composition and technique are continually advancing. It’s challenging, but that’s also the excitement.
Pairs trading strategy
To eliminate market risk and exploit temporary arbitrage opportunities from gaps in relative stock returns. Portfolios of correlated assets with similar market risk factor exposure using k-means, hierarchical clustering and affinity propagation clustering analysis.
S&P 500, FTSE 100, FWB, CAC40, MIL, SSE, HSI, TSE, ASX, TSX, BMV, BOVESPA
Hierarchical risk parity
Lopez De Prado’s hierarchical risk parity model – hierarchical clustering of stock returns’ covariance matrix to find a diversified weighting that distributes capital equally to each cluster hierarchy, using tree clustering, matrix seriation and recusive bisection
CAC 40, FTSE 100, FWB, HSI, MIL, S&P 500, SSE, TSX, TSE
Hierarchical equal risk contribution
Hierarchical tree clustering with matrix seriation and recursive division, but with naive risk parity within clusters. So for each cluster, the total cluster risk of an inverse-risk allocation is calculated.
ASX200, BMV, BOVESPA, CAC40, DAX, FTSE100, FWB, HSI, IT40, MIL, SSE, TAIEX
Dimensionality reduction
Principal component analysis to produce eigen portfolios. Dimensionality reduction reduces the instabilities in mean-variance analysis by producing linearly uncorrelated principal portfolios which directly lead to portfolio weights.
ASX200, BOVESPA, DJIA, CAC40, FTSE 100, HSI, IT40, PSE, TAIEX, TSX
Portfolio optimisation by genetic or evolutionary selection
The use of evolutionary algorithms to maximise portfolio Sharpe ratio. The process exploits elitism, carrying forward the best portfolio candidates as ‘parents’ for reproduction using tournament selection. crossover and mutation is generated tocreate the next generation of candidates. This step is repeated until the leading candidate is selected from the final populationwith portfolio weights that maximise the Sharpe ration criterion.
IDX, JSE, KOSPI, NSE, SSE, TSE
Portfolio optimisation by nested (hierarchical tree) clustering
Clustering assets based on correlations (distance matrices), finding optimal weights within each cluster, and then optimizing weights between clusters. Effectively divides large and potentially unstable covariance matrices into more manageable clusters, using dimensionality reduction.
ASX200, BMV, FTSE100, HSI, IT40, MIL, TSX, S&P500
Portfolio optimisation by Monte Carlo simulation
Efficient portfolio frontiers using Monte Carlo simulation, sequential least squares programming and convex optimisation
US tech, UK retail, Japan and Korea manufacturers, Canada and Brazil energy, Argentina utilities, India, China, Indonesia, Thailand & South Africa banking, Germany consumer, France luxury stocks
Yield curve level, slope and curvature simulation and prediction
Dimensionality reduction using neural networks. Supervised regression to forecast yield curve tenors
US Treasuries, UK gilts, bunds, JGBs, Canada, Australia, Singapore and Hong Kong bonds
Investor profile clustering with k-means and affinity propagation
Supervised models to predict investor risk tolerance used ina robo-advisor dashboard to improve portfolio management efficency
US investors
Stock and option pricing using Monte Carlo simulation
Least squares Monte Carlo simulations to overcome the ‘no closed form solution’ problem and develop approximations of the continuation value
US, Germany and Japan stocks
Portfolio multi-factor models
Using the Fama-French risk factor cross-sectional extension to CAPM
S&P 500 by industrial sector
Four- and five-factor extensions of the Fama-French CAPM
Incorporating historic returns and investment behaviour into the three-factor Fama-French model
Selected US, UK, Germany, Japan and China stocks
Stock return GARCH models and multi-variate volatility forecasting
Modelling stock return conditional volatility to manage volatility clustering with CCC-GARCH models
Selected US, UK, Germany, France, Italy, Spain, Japan, Korea, China, Brazil, Chile, India, Indonesia, Philippines, Mexico, South Africa, Canada, Australia stocks
Value at risk estimation
Estimating portfolio VaRs and CVaRs using Monte Carlo simulations
S&P, IT 40, Nikkei 225, SGX & ASX portfolios
Portfolio multi-factor models
Using the Fama-MacBeth risk factor cross-sectional approach
S&P 500 by industrial sector.
Registered office: 2 Villiers Court, 40 Upper Mulgrave Road, Cheam, Surrey SM2 7AJ
Registered in England 3333484
VAT Registration 711 1529 76
Duns 385640776