How to leverage AI in investments: Generali Asset Management's view

By Enrico Scarin, CFA , Head of Portfolio Solutions, ESG & Investment Data Science at Generali Asset Management

Artificial intelligence is revolutionizing asset management by improving process efficiency and the quality of investment signals.

Generali Investments is embracing these innovations, supporting various bottom-up projects and communities experimenting on various fronts. It is essential to focus on the specific processes of each company with a tailor-made approach that values the peculiarities of each reality.

We see great opportunities that require a culture of accepting mistakes as part of the path towards continuous improvement. AI is not just a tool but a catalyst for positive change and growth.

What is Generali Asset Management's vision on AI?

Artificial intelligence is a driver of innovation and sustainability. It allows us to offer personalized services, improving operational efficiency and risk management. Our asset management vision involves experimentation and adaptability, aware that errors and failures are an inevitable component of the journey.

We strongly believe that financial management should responsibly use these technologies, with ethics and transparency as fundamental pillars. We are committed to leveraging AI to create added value for customers and improve business operations.

What are the main initiatives Generali Asset Management is introducing in asset allocation?

Asset allocation is typically a quantitative discipline based on rigorous analysis of financial time series, such as expected returns and volatilities of various asset classes or their relative correlations. For example, in our investment committees, we have introduced internally developed clustering algorithms that use machine learning to improve financial forecasts and capital markets assumptions used in our studies.

This translates into more informed tactical decisions and more robust processes, with the possibility of building innovative portfolio strategies. It is an evolving process that allows complementing the portfolio manager's experience with a more data-driven approach. The results are promising and realistic in terms of performance, with risk indicators such as VaR, C-VaR, and expected drawdown significantly declining.

What are the objectives and expectations?

To grow, innovate, learn, and generate long-term value for our clients. Although some technologies are known, the necessary knowledge to improve processes can be found only within a company. Are our investment decisions informed? Can we apply our algorithms beyond asset allocation, or use LLMs for forecasts on time series with data quality issues? Can we create models on rates, stocks, and volatility? Or employ these tools in frontier markets or for illiquid assets?

We want to answer these questions in our day-to-day work, but it is now evident how these technologies are improving market forecasts in many companies, reducing errors in historical data analysis.

How do the new tools optimize the portfolio construction?

Professional portfolio construction combines top-down asset allocation elements with bottom-up market specialist expertise. If the analysis becomes more robust thanks to machine learning, management becomes more solid, and optimization between asset classes improves risk-adjusted returns.

In applying ML algorithms, we started from the idea of not formulating market hypotheses in the classical sense, rather using such expertise ex-post, after evaluating what the data tells us objectively.

Markets present cyclical elements, and these cycles show persistence elements that, when presented objectively, allow us to enhance the quality of our discussions, both on short-term estimates and for building medium-term portfolio strategies from a quantitative perspective.

Even at the bottom-up level, sentiment models fed by real-time data are increasingly used for selecting stocks or bonds, improving signal management.

These innovative approaches allow for higher returns and better risk management, demonstrating the importance of integrating advanced technology and specialist expertise.

What are the applications and benefits of AI in ESG portfolio analysis and allocation?

ESG analysis has become increasingly structured in recent years, making it ideal for AI support. Today, the analysis of unstructured data such as news, financial reports, and social media can be processed at unimaginable speeds until recently. AI algorithms can continuously identify and evaluate ESG risks, optimizing returns.

A portfolio can then be calibrated according to the investor's preferences, responding to specific needs.

For example, consider screening corporate financial statements, not only through numerical analysis but also by leveraging a machine's ability to read and massively analyze hundreds of texts in seconds and offer a brief summary of the ESG sentiment emerging from these data sources.

We have developed models both in the ESG field and to analyze emerging credit risks, topics that also present elements of contiguity. We will continue on this path, calibrating internal resources as best as possible, aware that business applications are truly numerous.

How do you position yourself in relation to the market and competitors?

Generali's asset management adopts AI at every point of the value chain.

We are at the forefront with AI and ML techniques, from deep learning neural networks in bond selection to clustering algorithms for sovereign ratings and strategic and tactical asset allocation techniques.

We create working groups that support portfolio managers, using Natural Language Processing to drastically reduce the production times of due diligence reports, extendable to other business areas.

In active management, we are close to the state of the art, with improvements on the horizon in terms of economies of scale.

Among the various initiatives already being implemented or in a very advanced stage of development, we also mention the creation of Quality Scores with ML support in stock selection.