Research

RAM AI’s researchers relentlessly explore new ways of extracting information from data to uncover new sources of return, increase diversification and improve liquidity.

Our conviction in the fundamentals and the non-linearity of their interactions has led us to develop a Deep Learning infrastructure, which helps reveal the predictive power of the exponentially growing volume of information at our disposal.

Using real financial news across various investment universes, our research team fine-tunes LLMs, training them to become experts at assessing the impact of newsflow on stock returns. The fine-tuned DeBERTa, Mistral and Llama models produce return predictions leading to strong portfolio performance, outperforming conventional sentiment models. Read full paper here.

Large Language Models (LLMs) have revolutionised Natural Language Processing and can be leveraged across several applications within the investment management industry. The below case study led by Tian GUO, PhD shows the benefits of modern LLMs in extracting information with significant predictive power on future stock returns, particularly for small and mid cap stocks. Read full paper here.

Nowadays, large language models are among the most impactful natural language processing techniques and have gained popularity in various applications, such as machine translation, language understanding, etc. These language models can be fine-tuned on domain-specific datasets to fit into the corresponding applications. In this paper, we will study the effect of financial news articles’ sentiments and text embeddings on predicting stock returns. Our in-house developed deep learning framework is able to model interactions between different input features and facilitates the comparison of feature combinations. Read full paper here.

In this paper we will look at the growing popularity of artificial intelligence and the potential for finance to prosper from machine learning implementations. As technology continues to push the boundaries of our imagination, new dimensions will undoubtedly emerge over time. Read full paper here.

Traditional market and factor datasets are typically structured in numerical terms in a form that can be digested by statistical models. Machine Learning can help deal with the abundance of rich textual data gleaned from financial news, earnings reports, and transcripts and their correlation to markets, currently, less exploited by quantitative managers. We take a dive into NLP techniques and their integration methods at RAM AI in the following Q&A. Read Q&A with RAM’s Data Scientist, Tian Guo.

Incorporating environmental, social, and governance (ESG) considerations into systematic investments have drawn numerous attention recently. In this paper, we focus on the ESG events in financial news flow and explore the predictive power of ESG related financial news on stock volatility. In particular, we develop a pipeline of ESG news extraction, news representations, and Bayesian inference of deep learning models. Experimental evaluation on real data and different markets demonstrates the superior predicting performance as well as the relation of high volatility prediction to stocks with potentially high risk and low return. It also shows the prospect of the proposed pipeline as a flexible predicting framework for various textual data and target variables. Read full paper here.

Flows into U.S. stocks dominated in 2024, driving concentration and valuations to record highs, U.S. equities representing a 72.5% share of the MSCI World Index—the highest on record. We examine historical parallels to uncover potential opportunities in this evolving market landscape. Read full paper here.

Following the unprecedented intraday VIX spike on August 5th, driven by panic hedging and the unwinding of short volatility positions by dealers, we analyse the role of the evolving volatility products landscape and explore the attractiveness of the volatility risk premium. Read full paper here.

At RAM AI, we have developed a deep learning framework to forecast stock returns based on a wide array of time-series inputs. This paper illustrates how shorter-term strategies have attractive standalone characteristics and benefit lower-frequency books.  Read full paper here.

The profits a firm generates can either be distributed to its shareholders through dividends/share buybacks, pay back some of its debt or be used to invest. A company invests in its future organic growth either through Research and Development (R&D) or by engaging in Capital Expenditures (Capex). In a previous research paper, we studied the effect of Capex announcements on stock returns. The present document analyses R&D activity and its impact on company fundamentals. Read full paper here.

The profits a firm generates can either be distributed to its shareholders through dividends/share buybacks, reduce its debt or be used to invest. A company invests in its future organic growth either through Research and Development or by engaging in Capital Expenditures (Capex). As of January 2021, the net Capital Expenditure on sales ratio reached 2.54% for US companies. Read full paper here. 

ClimateControversyBERT is an open-source language model developed to detect and classify corporate climate controversies in real time, such as brown projects, greenwashing, or ambiguous actions. In this paper, we analyse the market’s reaction to these controversies, showing significant negative impacts—especially for firms with existing climate commitments. Read full paper here.

The Sustainable Finance Disclosure Regulation (SFDR) is a European regulation introduced in 2019 and applicable from 10th March 2021 to improve transparency in the market for sustainable investment products and in sustainability claims made by financial market participants, and prevent greenwashing. In this paper, we analyse the impact an SFDR governance framework can have on return and style biases. Read full paper here.

Over recent years, data measuring firms’ sustainability characteristics has proliferated. An important challenge that investment managers encounter is the combination of ESG and traditional factors such as value, growth and momentum, which is crucial for an optimal stock selection and portfolio construction process. RAM AI’s systematic equity team has developed a deep learning framework that models the interaction among different input features. In this paper, we will evaluate the efficacy of this framework when tuned to combine ESG and traditional factors for a stock return prediction task. Read full paper here.

Incorporating climate considerations into portfolio analysis and systematic investments has drawn numerous attention recently. It is motivated by the pursuit of sustainable investing for a low-carbon transition. In this paper, we propose to integrate both structured and unstructured climate-related data into quantitative investing for stock markets, e.g. carbon emission scores and climate events from news flows. We develop a deep learning framework to consume these data for assessing climate-related opportunities and the risk of stocks in the investing universe. Experimental evaluation on real data demonstrates the low-carbon intensity of the constructed portfolio as well as decent investing return. Read full paper here.

The rapid growth of ESG data in recent years makes it a compelling new dimension to the investment process, but integrating ESG presents a wide array of challenges, from noise filtering to hidden biases. In this paper we introduce some of the shortcomings of ESG data, as well as the value-added ESG inputs can bring into a stock selection process.   Read full paper here.