Portfolio Optimization in the Age of Big Data
Machine learning offers exiting new opportunities for financial portfolio optimization. VDG Analytics has undertaken a study to examine the potential synergies between neural networks and the Black-Litterman model when creating optimized portfolios of index funds.
Machine Learning and Black-Litterman Optimization
Seeking to examine the opportunities provided by combining machine learning with advanced financial models, VDG Analytics has constructed a Long Short-Term Memory Recurrent Neural Network leveraging machine learning in synergy with the Black-Litterman model to refine the optimization of index fund portfolios. We compared performance outcomes of portfolios optimized via the neural network against those shaped by the Capital Asset Pricing Model and the Mean-Variance approach. The comparison reveals that while the neural network may not craft superior portfolios by traditional benchmarks, it demonstrates an ability to predict future excess returns with greater reliability.
See the full report for more information about the results of this project and the methodology used in creation of the neural network.
Long-Short Term Memory Recurrent Neural Network
The use of machine learning in financial modeling was at the core of this project. While many forms of machine learning could have been used, we chose to build af Recurrent Neural Network using Long-Short Term Memory.
This type of neural network is particularly suitable for financial time series due to its ability to remember long-term dependencies, which is crucial for capturing the patterns in market data. Our model uses the Keras deep learning framework and is composed of two 50-neuron layers, designed to grasp the fluctuations of financial data, discerning optimal market positions from historical returns and volume.
Once trained, the model is capable of making sophisticated forecasts that can inform investment decisions, rather than simple trend extensions. These forecasts can then be integrated with the Black-Litterman model.
The Black-Litterman Model
While the use of machine learning in investment and portfolio optimization has quickly become widespread among those that have the capacity to use it, at VDG Analytics we have seen noticeably less interest in the combination of machine learning with advanced financial models. We wanted to understand if it was viable to combine these approaches to optimization, so we chose a model that specifically allows for the addition of investor views, the Black-Litterman model.
The Black-Litterman model is a method for quantitative portfolio optimization that blends market equilibrium-based predictions from the widely used Capital Asset Pricing Model with individual investor views, resulting in customized, risk-adjusted asset allocations.
The advantage of using this model is that instead of relying purely on historic market data or on investor views, it combines the two with a complex set of calculations, balancing the confidence in the views with the reliability of the market trends. Using this model, we where able to make our neural network generate views about the market and plug them into the Black-Litterman model as investor views. Our theory was that tempering the neural networks predictions this way would lead to more grounded and reliable portfolios.