AI meets Value Investing
The (r)evolution
Since the 1920s, the way in which undervalued companies are identified has changed fundamentally. From the purely manual approach of the analysts of days gone by, to the computerized support of the 1980s, to today’s self-learning AI systems – analysis has entered a new era.
Our AI-based solution is revolutionizing fundamental analysis by processing huge amounts of data in real time and recognizing patterns on its own. While humans set the targets, artificial intelligence takes over strategy and analysis – faster, more precise, and free of human bias. Immerse yourself with us in the future of investment analysis and experience how AI is paving the way to new insights.
— Since the 1920s —
Manual process
Identification of undervalued companies through manual fundamental analysis
| Objective: | Human |
| Strategy: | Human |
| Analysis: | Human |
— Since the 1980s —
Automated process
Identification of undervalued companies through computer-based fundamental analysis
| Objective: | Human |
| Strategy: | Human |
| Analysis: | Machine |
— Since the 2010s —
Self-learning process (AI)
Identification of undervalued companies through self-learning fundamental analysis
| Objective: | Human |
| Strategy: | Machine |
| Analysis: | Machine |
Quant vs AI
A quantitative investment approach typically relies on predefined, human-created rules that are often derived from linear data analysis. These rules are applied statically and are usually optimized for a specific point in time. Frequently, factor investments focus on a single dominant key indicator, limiting their adaptability to changing market conditions.
In contrast, an artificial intelligence model autonomously uncovers complex, non-linear correlations and continuously adapts to evolving market dynamics through adaptive self-learning—all without forgetting past events. Here, humans do not define the rules; instead, the model learns them. Non-linearities are crucial for shaping the model, enabling diverse key indicators to be interconnected in virtually limitless combinations.
Old quant world
- ■ Model building by humans
- ■ Model based on a priori views on how markets work
- ■ Convictions / drivers are implemented
- ■ Implementation of ‘if-then’ scenarios
- ■ Backtesting – look-ahead bias
- ■ Fixing the model and going live with it
Static rule-based strategies
Static, rule-based strategies typically focus on targeting specific drivers of returns, known as factors, to enhance performance and manage risk. In the quantitative investment space, this approach uses data analysis and statistical models to identify and capitalize on these factors. Commonly used factors include value, growth, momentum, size, quality, and volatility.
The importance of each factor is determined by humans, who, for example, decide the weighting of each input feature for the factor—illustrated by the thin grey lines connecting input features to grey boxes. Thicker lines indicate a greater influence of the feature on the factor; for instance, dividends have a larger contribution to the value factor than sales multiples. Once set, these dependencies remain fixed.
Quantitative investors leverage historical data to create rules-based models that screen and select securities exhibiting desirable factor traits. This approach aims to outperform traditional indices by systematically exploiting the risk premia associated with each factor.
For a short, readable, high-level overview, see Robeco’s “Guide to Factor Investing in Equity Markets“. If you are interested in going deeper, we recommend Ilmanen’s masterpiece “Expected Returns: An Investor’s Guide to Harvesting Market Rewards“.
New AI world
- ■ Model building through AI
- ■ Model not based on a priori views of how markets work
- ■ Extracting features from data
- ■ Create model from data
- ■ Walk-forward testing – no look-ahead bias
- ■ Model learns and adapts continuosly
Adaptive self-learning startegies
AI-driven investment strategies—illustrated here as a neural network—do not rely on human-defined factors. Instead, the model automatically learns which input features are important by adjusting weights during training. It first takes input data and makes a prediction through a process called forward propagation, then compares this prediction to the correct answer to measure the error. The model then works backwards to fine-tune the connections (weights) within the network—this step is known as backpropagation. This two-way flow of information through forward and backward propagation allows the network to learn from its mistakes and become more accurate over time.
Neural networks excel at detecting non-linear relationships and hidden patterns that traditional factor models might overlook. The network’s architecture enables interconnected features (e.g., Feature A is connected to all input features), allowing the model to infer interdependencies. For instance, AI can simultaneously consider multiple input features—such as sales, profit, growth, and market share—and understand their combined influence on asset prices.
However, AI models can be more challenging to interpret and require significant computational resources.
The Future of AI-Powered Investing
By combining Buffett’s investment principles with modern AI technology, we aim to revolutionize the way investors evaluate stocks.
The Vision
Imagine having the sharp mind and unrivaled market knowledge of Warren Buffett – in every single sector. With our innovative AI, we analyze companies from a wide range of industries with the precision and expertise of an experienced investor. Whether in energy, retail, food processing or manufacturing, our AI brings Buffett’s principles to every industry and discovers the companies with the best growth opportunities.
Experience how Artificial Intelligence applies the know-how and analytical skills of a “Warren Buffett” to any sector to identify undervalued opportunities. Make data-driven decisions and invest smarter – with AI that thinks like one of the best investors of our time.
The Implementation
How our AI stock picker works:
Zoom into single industries such as technology, healthcare of energy and construct all possible company pairings.
Using historical time-series data of fundamental company metrics (e.g., revenue, earnings, debt etc.), the model learns to distinguish companies with the highest long-term growth potential.
Compairing all company pairings yields a ranked list of most promising investments per sector. Should you be more interested in the idea of how ranking works, go back to the roots via the paper Learning to Rank: From Pairwise Approach to Listwise Approach by Cao and Qin.
smart investment
Our global AI-Buffett-Index
Since May 1, 2022, we have been putting this idea into practice: artificial intelligence that brings the smart investment approach of Warren Buffett to every economic sector. Our AI analyzes thousands of companies every quarter and identifies the 50 most promising investments from various sectors – companies with strong fundamentals and the potential for above-average growth.
This exclusive portfolio is calculated as an index at Solactive. With the Sector AI Buffett Index, we invest in a data-based and broadly diversified manner in the best companies recommended by our AI analyses – an intelligent strategy that combines proven investment principles with modern technology.
Performance
Market capitalization
Most members of the ACATIS AI Index are medium-sized companies, i.e. in the range between €1 billion and €10 billion.
This is no surprise as most of the investable universe falls within this interval. As explained in the article “The market size dilemma”, small and micro caps in particular are excluded. The large caps segment comprises companies with a value of more than €10 billion, but still less than €100 billion.
Sector weights
As a business picker, our approach aims to find the best companies within a particular sector, no sector allocation technique is used.
Consequently, we replicate the sector distribution of global developed equity markets as closely as possible, i.e. we create the index sector neutral. However, we exclude the financial and real estate sectors and are controversial with regard to the biotechnology industry. Why is that? See our article “Why exclude financial and real estate companies?”.
Books on machine learning in the financial sector
There is not a large, high-quality selection of literature in this area. However, three books can be recommended without hesitation.
Collection of great papers about ML for asset management
Academic work on artificial intelligence in the financial sector
There is an almost unmanageable number of articles about using ML for investing (and especially for trading). Unfortunately, among them there are also many works of moderate quality. But of course there are also papers with outstanding content. Below you will find a short list with selected papers.
Disclaimer: Some of the articles might be paywalled
Financial Machine Learning (2023)
by B. Kelly, D. Xiu
A survey of literature on machine learning in the study of financial markets
Uncertainty-Aware Lookahead Factor Models for Quantitative Investing (2020)
by L. Chauhan, J. Alberg and Z. Lipton
Absolute favorite paper on applying ML to asset management in the context of predicting future fundamentals. Add-on by incorporating uncertainty estimates. That’s how you do it!
Can Machines Learn Finance?
by R. Israel, B. Kelly and T. Moskowitz
… or what are the challenges of applying machine learning to finance: 1) Low signal-to-noise ratio, lack of data, market regime switch alias non-stationarity.
Missing Financial Data (2023)
by S. Bryzgalova et al.
Financial data are challenged by missing values. This work documents the widespread nature and structure of missing observations of firm fundamentals and show how to systematically deal with them.
Deep Learning for Global Tactical Asset Allocation (2018)
by G. Chakravorty et al.
Although tactical asset allocation is not our primary interest, a recommendable paper that stands due to its clean and neat methology.
The 10 Reasons Most Machine Learning Funds Fail (2018)
by Lopez de Prado
Marcos López de Prado is currently one of the hotest names in the field. This paper collects some common pitfalls and solutions when applying machine learning to investing or trading. As a scientist, I am sometimes surprised that references to meticulous work and adherence to basic statistics can generate such a hype.
Autoencoder Asset Pricing Models (2019)
by S. Gu, B. Kelly and D. Xiu
Simply a nice idea to use an autoencoder for asset pricing, and AQR is not the worst address either.
When humans try to outsmart machines! The growing number of AI readers is motivating companies to write their documents in such a way that they can be better analyzed and processed by machines. Companies avoid words that are perceived as negative by computer algorithms compared to words that are perceived as negative only by dictionaries for human readers. This paper examines the feedback effect on corporate disclosure in response to the use of technology.