Fundamental investment success involves developing a view on global trends and future market directions as well as identifying relevant investments that are aligned with this strategy. This second step of the process is long and labourious.
Previously, applying machine learning to assist would require the use of scientists with little knowledge of investment. Their role would be to review financial data of potentially thousands of relevant businesses and identify under- or over-priced investments. Highly accessible machine learning platforms are transforming the process.
• It is simple to get started. An investment analyst extracts relevant financial data – anything from revenue and earnings per share (EPS) to operating margins – from a financial research platform such as Bloomberg. Once a suitably sized dataset has been acquired, an investment analyst then awards binary scores of one to good investments, and zero to inadvisable investments. These can, for example, be based on whether the stock price increased over the past year or not.
• Data preparation is typically a time-consuming task for qualified data scientists – with many stating it is the most time consuming and least enjoyable part of their work. With a humanised machine learning platform, powerful data preparation and manipulation capabilities are made accessible to employees of all skill levels – in this case, an investment analyst – cutting the time and resources required to prepare financial data.
• A key function of such platforms is providing actionable advice on how to best correct errors within datasets. In an investment context, take a dataset containing missing values for share payout ratios. A humanised machine learning platform will flag this issue and provide several means to correct it. The investment analyst may have contextualised knowledge of why this is the case – such as no dividends being issued – and take the opportunity to automatically fill in all missing payout ratio values with zero.
To save time on similar activities in the future, the investment analyst can save the data preparation workflow for auditing and reuse purposes by colleagues.
At this stage, it is possible to view a visualisation of how promising investment targets are distributed based on captured financial data points. It is a strong possibility t no clear pattern will be evident. However, user-led data preparation is an invaluable tool in providing investment analysts with an opportunity to visualise and manipulate raw data prior to model selection and deployment.
• Intuitive machine learning platforms identify an effective model with accuracy to apply to datasets and provide justification for the suggestion. These models avoid overfitting – that is, fitting models to training data to the extent that they struggle with unseen data and fail to provide accurate predictions.
One approach is for the platform to use Bayesian statistics which learn from each iteration what works and what does not to quickly identity the most effective model with a high forecasting accuracy. Selected models can be used within the platform or deployed through an application or Excel spreadsheet to make live forecasts.
• Machine learning models typically uncover intricate relationships between complex financial data points.
An investment analyst can go further and ask their machine learning platform to cluster. This can then be used for final risk analysis when constructing an investment portfolio.
• Humanised machine learning platforms offer highly intuitive, user-centric interfaces to guide users from data preparation to model deployment. An investment analyst can directly use machine learning to unlock financial data, at speed and scale, without training or data science expertise. The best way to get started is to pick a stage of the investment process which could benefit from experience gathering. This will enhance the analyst’s ability to perceive experience and as a result improve the performance of the portfolio by adding one more step in the decision-making process. At this stage, the investment analyst has become a citizen data scientist.
These capabilities also offer promising applications in the financial sector beyond informing investment analyst strategies. Strong use cases today include using machine learning to forecast company revenues to derive valuations, predict the failure of trades for trade settlement, predict and anticipate large drawdowns, forecast EPS beats and misses, and optimise marketing and sales operations for funds that rely on distribution strategies.
Charles Brecque is CX operations manager at University of Oxford machine learning spin-out Mind Foundry