Here’s How Machine Learning is Changing the Investment Process

Here Martijn Groot, VP Marketing and Strategy at Asset Control , discusses how AI and Machine Learning techniques are finding their way into financial services and changing the way we do things, in particular how we invest. Ranging from operational efficiencies to more effective detection of fraud and money-laundering, firms are embracing techniques that find patterns, […]

Here Martijn Groot, VP Marketing and Strategy at Asset Control , discusses how AI and Machine Learning techniques are finding their way into financial services and changing the way we do things, in particular how we invest.

Ranging from operational efficiencies to more effective detection of fraud and money-laundering, firms are embracing techniques that find patterns, learn from them and can subsequently act on signals coming out of large volumes of data. The most promising, and potentially lucrative, use cases are in investment management though.

Among the groups that benefit most are hedge fund managers and other active investors who increasingly rely on AI and machine learning to analyse large data sets for actionable signals that support a faster; better-informed decision-making process. Helping this trend is the increased availability of data sets that provide additional colour and that complement The typical market data feeds from aggregators, such as Bloomberg or Refinitiv, range from data gathered through web scraping, textual analysis of news, social media and earnings calls. Data is also gathered through transactional information from credit card data, email receipts and point of sale (“POS”) data.

The ability to analyse data has progressed to apply natural language processing (NLP) to earning call transcripts to assess whether the tone of the CEO or CFO being interviewed is positive or negative.

The ability to analyse data has progressed to apply natural language processing (NLP) to earning call transcripts to assess whether the tone of the CEO or CFO being interviewed is positive or negative.

Revenue can be estimated from transactional information to gauge a company’s financials ahead of official earnings announcements and with potentially greater accuracy than analyst forecasts. If, based on this analysis, a fund believes the next reported earnings are going to materially differ from the consensus analyst forecast, it can act on this. Satellite information on crops and weather forecasts can help predicting commodity prices.

These are just a few examples of the data sets available. The variety in structure and volume of data now available is such that analysing it using traditional techniques is becoming increasingly unrealistic. Moreover, some has a limited shelf life and can quickly become out-of-date.

Scoping the Challenge

Setting up a properly resourced team to assess and process this type of data is costly.

The best approach therefore is to more effectively assess and prepare the data for machine learning so that the algorithms can get to work quickly. Data scientists can then focus on analysis rather than data preparation. Part of that process is feature engineering, essentially selecting the aspects of the data to feed to a machine learning algorithm. This curation process involves selecting the relevant dimensions of the data, discarding for instance redundant data sets or constant parameters, and plugging gaps in the data where needed.

An active manager could potentially analyse hundreds of data sets per year; the procedure to analyse and onboard new data should be cost-effective. It should also have a quick turnaround time as the shelf life of some of these data sets is short.

Addressing these challenges means that traditional data management (the structured processes to ingest, integrate, quality-proof and distribute information) has to evolve. It needs to extend data ingestion and managing data quality into a more sophisticated cross-referencing of feeds looking for gaps in the data; implausible movements and inconsistency between two feeds. For instance, speed of data loading is becoming more important as volumes increase. With much of the data unstructured, hedge funds should be conscious of needing to do more with the data to make it usable. More sophisticated data mastering will also be key in making machine learning work effectively for hedge funds.

This functionality coupled with the capability to quickly onboard new data sets for machine learning will enable funds to save money and especially time in the data analysis process. It will allow data scientists to focus on what they do best and generate more actionable insight for the investment professionals.

Reaping the Rewards

Machine learning clearly has huge potential to bring a raft of benefits to hedge funds, both in reducing the time and cost of the data analysis process and in driving faster time to insight. It also gives firms the opportunity to achieve differentiation and business advantage. Hedge funds need to show returns to attract investment in an increasingly competitive space, machine learning supported by high quality data management offers a positive way forward.

Leave A Reply