Unlike traditional models, machine learning can simulate millions of potential portfolio allocations. It can also assess which allocations perform best under different market conditions while strictly meeting investor risk preferences.
ML algorithms can continually adjust portfolio weights to maintain optimality for the investor as volatility, correlations and macro conditions evolve. Reallocations can be automatic or at the fund manager’s discretion, or a combination of both.
ML also lends itself to multiple scenario analysis. ML models can simulate market shocks, oil price shocks, wars and changing tariffs, for instance, and stress test to measure portfolio resistance and to satisfy regulators.
Deep learning ML models can predict short- and long-term risk across and between asset classes. They can monitor the market state and classify, for example calm, agitated or crisis, and adjust forecasts accordingly.
ML models have powerful capabilities in correlation analysis, identifying dependencies and relationships between assets that more traditional analytical models might miss. They can also monitor how these relationships might be evolving and simulate future development.
ML models are ideally suited to detecting market anomalies. Instances of unusual trading, sudden changes in market liquidity or outright fraud can immediately be brought to the attention of the portfolio manager.
AI has transformed macro forecasting in fund management in recent years. Economic indicators, global trends, policy decisions and market sentiment can be measured in real time to optimise asset allocation and risk exposure.
ML models can integrate data from multiple sources and different formats into concise, interpretable statements, freeing time, effort and resources for the portfolio manager.
Deep learning models can continually monitor for fundamental regime change – to alert the portfolio manager when markets have moved from one state to another, for example from negative to positive inflation-adjusted yields.
ML models can analyse enormous datasets from a variety of diverse sources and formats, for example from earnings reports, social media and satellite imagery. They detect connections and identify potentially underpriced assets that more traditional models miss.
Natural language processing models can be used to compile news, price movements and viral messages to calibrate market sentiment and extract forward-looking information.
Underpriced assets can be found by using non-traditional data sources such as weather forecasts, foot traffic at retailers and union strike votes.