Decision-making in today's disrupted and increasingly complicated environment needs to be contextual, continuous and connected for positive results. As per a Gartner survey, 65% of decisions made today involve more stakeholders and choices, making them more challenging than they were in the past. AI and big data have the potential to augment decision-making, in a scalable, adaptive and personalised way. As computing power has grown and the world generates previously unimaginable volumes of data, the potential of AI models has grown.
Processing vast amounts of data within seconds, big data and machine learning (ML) are empowering companies to make strategic decisions in an agile manner. This includes making strategic investments. ML can enhance how managers implement strategies and fine-tune portfolio allocation. Even the most fundamental and non-quantitative managers can generate ideas from ML-sourced and synthesised data.
For example, unstructured data from diverse datasets like FCA filings, company earnings calls, websites, and social media can be used to extract company performance information and market sentiment. Machine learning makes it possible to predict performance and identify high-growth stocks through applying ML enhanced dividend forecast data. With so many use cases, the global AI in asset management market size is expected to reach $13.43 billion by 2027.
The Benefits of AI and ML in Investment Decisions
While quantitative investors might have access to real-time data, they might not have access to an organised dataset that can be analysed to provide investment ideas. Within AI, an important subfield for the investment world is ML, through sophisticated algorithms that can explore data with limited human intervention. AI, when applied in this way can deliver more accurate forecasts and insights. For instance, it can automate market sentiment analysis, deepen trader risk profiling, drive news and event analytics, and even reputational risk management. Let’s look at an example of how ML can drive dividend forecasting for a company.
An ML model is designed to answer a query based on specific input data. Consider a question like, “How much will company ABC increase its dividend in 2022?” Or “By how much do we predict company XYZ will reduce its dividend amount in 2022?”
The model is trained with thousands of examples, using historical actual data. Based on this data (the training data) predictions of interest arise for further research. In this case, the training set consists of large dividend increases (for instance, special dividends) and decreases (such as, reductions), industry trends, competition, and company financial history for a large number of securities globally over 5 years. With the increased availability of large financial datasets, the system can identify underlying patterns – given time and experimentation.
The development of a robust ML model involves several iterations and trial and error. We continue to experiment with various datasets and input fields to feed the model the most relevant data for increasingly accurate insights. The system analyses data via multiple layers of learning, starting with the basic concepts and then combining these to answer complex questions. The use of artificial neural networks means that data is passed through multiple layers of non-linear processing units; mimicking the way neurons behave in the human brain.
Overall, the deployment of AI/ML in investment management companies can drive advantages through 2 primary avenues:
- Higher Efficiency: AI/ML can drive competitive advantages for financial services companies via cost reduction and increased productivity. This can be made possible through enhanced decision-making processes, improvements in risk management, automated execution, regulatory compliance, and more.
- Improved Results: By customising products and services and launching new product offerings, companies can literally improve returns for themselves / investors.
The AI Disadvantage for Investment Decisions
Smart machines lack the emotional intelligence of human beings, which might play an important role in specific situations. For example, AI and machine learning are fail at the task of valuing a home. The example of online real estate giant, Zillow Group, is a clear example. The company recently decided to close down its home-buying business, leading to massive job losses and stock price decline. Co-Founder and CEO Rich Barton attributed this closure to the company’s lack of confidence in its proprietary home-buying algorithm to accurately forecast home price fluctuations.
“But what we can’t solve is what the model is going to tell us about how much capital we need to raise, deploy and risk in the future in order to achieve a scale that we think is necessary to offer a fair price to customers for their homes in a competitive way,” he said.
The case of Zillow indicates that companies can’t rely solely on algorithms for decision-making. For example, in 2019, the use of facial recognition was banned by San Francisco legislators, as they showed proneness to errors when it came to users with dark skin. So, when it comes to an organisation deploying AI, there have to be processes to mitigate bias.
At Woodseer, AI/ML augments human capabilities, rather than replacing it. Our ML model makes our dividend forecast data more accurate – in particular with reference to otherwise un-anticipated outliers. The ML models provide suggestions for specific review by our UK and Canada based analysts.
The Best of Human and Machine Intelligence Combined
Much of what ML models are expected to do is ultimately human behaviour, but without human experience and understanding. They are only trained to work with the “relevant” data provided, and the word “relevant” here is crucial. We can build up a comprehensive outline of a company’s health, based on financial reports, media coverage, industry trends, and more, but at the basic level, a company consists of individuals working in a dynamic world. Their experience, motivations, and aspirations can’t be captured in a database. Dividend payments (and dates) are ultimately at the discretion of company executives, working with limited information and personal biases.
At Woodseer, our aim is to provide cutting-edge technology, together with the strength and expertise of an experienced and skilled team of analysts. We implemented our hybrid algorithm+analyst approach for dividend forecasting since 2017 for every publicly-listed company in the world (plus ADRs and ETFs). We emphasise human primacy in decision making, in higher value use cases, focusing on 3 key areas:
- Manual reviewing of estimates upon not meeting certain criteria
- Scheduled reviewing of anticipated dividend pay-outs and important securities
- Addressing client queries regarding specific predictions.
This hybrid model delivers accuracy at scale. Over 80% of our forecast numbers are exact or within 10%, and 80% of forecasted ex-div dates are exact, or within 7 days.