It seems that automation has already eliminated the fund manager. Investors have now been convinced that fee margins approaching or even exceeding 1% are unlikely to be recouped through better decisions. Assets under management (AUM) of passive investment ETFs now make up 72% of the North American market, according to data from Bloomberg.

Now, with the explosive advancement of AI technology since the release of ChatGPT late last year, it might be tempting to conclude that the rest of the gap will be closed shortly. AI was one of the hot topics at the World Investor Week (WIW) forum, held October 2-6 at TABF. Several aspects of international fund management were analyzed by various speakers, including the rise of active ETFs, market segmentation by investment theme and strategy, and ESG investing, as well as the emerging role of AI in the industry.

Several AI-managed ETFs have already appeared on the market, but their application remains limited for now. In fact, AI is not the only replacement for high-cost funds, and that may not even be the best use case for AI in fund investing.

ChatGPT is not a mathematician

The type of automation performed by AI (or more precisely, deep learning) is very different from the automatic indexing of classic ETFs, or even the sophisticated mathematics performed by ‘quant’ investors. One frequent misunderstanding of its novelty involves so-called “dimensionality reduction.” In basic statistics, one asks a (hopefully well-specified) question and then receives back an answer. This approach can however fail to discover and utilize the hidden patterns that may exist in the data, so sometimes it is also necessary to “ask” the data what questions one should be asking in the first place. Therefore, principal components analysis (PCA) has been used for a long time, including quant investing as well as a variety of scientific and other fields.

Nevertheless, in fund management, humans must still vet potential correlations of interest, and most importantly to decide when existing relationships no longer apply – after all, every value-producing trade is eventually discovered by other investors. This judgment process can be ad hoc and difficult to verbalize, making it a tough target for automation by AI.

The true novelty of AI, on the other hand, is not its ability to output more useful information than its input, but rather to ingest and integrate almost any kind of data without restriction. Earlier statistical techniques tended to work on structured data, like market prices, but could not interpret unstructured inputs, like natural language or videos, without the use of pre-processing techniques which would strip away some portion of the useful information. This makes AI a qualitative, rather than a quantitative investor. A full matrix of investment automation styles can thus be formed, as shown in Fig.

According to analysis presented by Prof. Shih-Chuan Tsai of National Taiwan Normal University at the conference, the AIEQ and AIIQ funds have not outperformed the market overall since they were launched in 2017 and 2018, respectively. Nevertheless, Tsai still identified two areas of superior performance: international investing and market shorting. The former requires broad analysis of investment conditions in diverse environments, all for relatively small portfolio positions – in other words, a perfect target for outsourcing. The reason for the success of the latter, meanwhile, is somewhat more subtle. The success of short strategies is typically driven less by market fundamentals than by investor psychology, which can be understood by integrating a wide variety of information sources, including non-traditional ones, a similarly labor-intensive task.

Environmental protection is a qualitative problem

The most convincing use case of AI in investing may not be to cut costs directly, but rather to drive further value. One promising application is ESG, which is not a single alternative dimension to financial returns, but rather a collection of measures with unclear and possibly changing relationships.

Current challenges in ESG implementation include a lack of standardization between systems, a lack of input data for calculations, a need to quickly educate SME borrowers about sustainability, and human resource challenges. These are precisely the sorts of problems at which AI excels – not for making decisions directly, but at organizing diverse information into human-readable insights.

Based on remarks from several speakers at the conference, it is important not to approach ESG from a compliance perspective. Rather, after simply implementing it, the compliance aspects will tend to fall into place.

This wisdom can also be applied to AI assistance. One of the most vexing limitations of AI to date has been “explainability,” or transparency into decisions by models, which makes it difficult to apply such models in highly regulated areas. Investors generally hope for their portfolio allocation decisions not to be a black box; AI is worryingly prone to hallucinations, or conclusions which “sound like” facts, rather than reflecting real facts. It is therefore easier to envision a role of AI in flagging interesting patterns or emerging trends for humans to follow up on.

Humanity strikes back

It is also possible to question the premise that AI will be able to find use cases in the first place. An alternative possibility is that passive investment as a whole gets rejected in favor of a new generation of active investment, but this time hopefully at lower cost – the upper right quadrant of the matrix. The value of the global pool of investment savings is in the multiple tens of trillions of US dollars. Shouldn’t humans be actively involved in allocating that money?

Merryn Somerset Webb argued in a recent Bloomberg column that passive investment allocation by market capitalization is in fact no different from momentum investing: trending assets are disproportionately selected. Such a strategy may have worked well in an inflationary, debt-constrained environment, but going forward, it will become more important to weed out overvalued investments and seek out the hidden gems. It is possible that overall market exposure alone may not provide the same returns and risk resilience that it has over the past decade. Active ETFs therefore provide a happy medium, including some of the tax advantages of passive ETFs, and are growing quickly in Europe.

Featuring complex systems and highly consequential decisions, financial markets are a rich area in which to demonstrate the strengths and weaknesses of AI. It has been called a “general-purpose technology” like electricity or the internet, a new driver of productivity with the potential to drive innovation in a wide variety of fields for the foreseeable future. That designation however does not mean that it can be applied equally to every sort of problem, as implied by the narrative of human replacement. The coming process of market integration will involve discovering where it indeed has strengths, and where it must be complemented with human labor.