Artificial intelligence (AI) evangelists promise a revolution for the economy. However, deploying AI to gain an edge in financial markets and generate excess investment returns is increasingly in doubt.<\/p>\n
A paper published by the International Monetary Fund in November 2023 said financial sector spending on AI is projected to more than double to US$97 billion by 2027, rising at a 29% compound annual growth rate.<\/p>\n
And recent research from Statistics Canada singled out financial sector professionals as the cohort most exposed to AI disruption, alongside computer programmers and IT professionals.<\/p>\n
\u201cA broader segment of the labour force could be affected in an era when sophisticated large language models such as ChatGPT increasingly excel at performing non-routine and cognitive tasks typically done by highly skilled workers,\u201d StatCan\u2019s paper said.<\/p>\n
Even as the financial sector appears ripe for AI disruption, the feasibility of companies replacing human workers with AI-powered technology is unclear. StatCan noted there are legal, financial and institutional barriers to replacing educated professionals with tech.<\/p>\n
Further, the promised payoff from AI for financial firms will probably be limited.<\/p>\n
In a report, analysts with Moody\u2019s Investors Service Inc. point out that \u201charnessing AI is fraught with technical and organizational challenges.\u201d<\/p>\n
While the technology can automate routine tasks and enable portfolio managers to process data \u2014 and potentially glean novel investment insights \u2014 practical constraints mean AI deployment is unlikely to drive vastly improved investment performance.<\/p>\n
According to the Moody\u2019s report, generative AI tools aren\u2019t particularly useful for uncovering investment signals.<\/p>\n
\u201cUsing large language models to identify investment opportunities is tempting, but this approach faces several problems,\u201d Moody\u2019s said.<\/p>\n
For example, large language models may be great at processing large volumes of text, but they aren\u2019t particularly good at assimilating financial data. They\u2019re also prone to \u201challucinations\u201d \u2014 producing inaccurate outputs or identifying non-existent patterns. And slight variations in inputs, or in the models themselves, can produce wildly different results.<\/p>\n
Some of these issues are documented in research from Desjardins Group, which compared inflation forecasts made by human economists with forecasts from generative AI tools.<\/p>\n
While the models performed well in a small sample, their forecasts tended to be \u201cmore momentum-driven\u201d than economists\u2019 forecasts and failed to \u201caccount for idiosyncratic behaviour in some subcategories of inflation in the face of very specific shocks, such as expected tax hikes or subsidies to government programs.\u201d<\/p>\n
Some models generated forecasts that just repeated the outcomes of the previous period. \u201cThis also speaks to the limited transparency in AI model data, training and forecasting processes,\u201d Desjardins said.<\/p>\n
Additionally, efforts to have the AI models generate historical forecasts were often thwarted by the models using information unavailable at the time, despite being instructed not to do so, Desjardins\u2019 report noted.<\/p>\n
Ultimately, the report concluded, \u201cAI isn\u2019t ready to replace professional analysts yet, particularly in smaller, less information-rich economies like Canada.\u201d<\/p>\n
While the expectation is that AI models will improve at forecasting over time, Moody\u2019s said the availability of these tools means the answers they produce are likely to have limited value in generating higher returns. To the extent that investors use the same advanced tools to evaluate the same securities, the value of any particular investment insight these models produce will be quickly assimilated into the market \u2014 leaving individual users with no advantage.<\/p>\n
To gain an edge, firms will probably need to run their more traditional AI models, Moody\u2019s suggested.<\/p>\n
Traditional AI models are more useful than large language models for processing large sets of economic or financial data, Desjardins\u2019 report said, but they are much harder to use \u201cbecause they require investors to train and maintain the models themselves.\u201d<\/p>\n
In the face of these challenges, most asset managers haven\u2019t implemented such solutions, \u201coften struggling to scale beyond proof-of-concept stages,\u201d Moody\u2019s said.<\/p>\n
Even if proprietary models get off the ground, they will have to prove reliable and understandable enough for analysts and portfolio managers to rely on them.<\/p>\n
\u201cAsset managers will need to set up robust systems to monitor when a model\u2019s investment strategy stops working and decide whether it can be retrained or should be discontinued,\u201d Moody\u2019s said.<\/p>\n
Still, asset managers can\u2019t ignore the technology\u2019s potential to shake up the investment business and other sectors. Firms unprepared to take advantage of AI risk falling behind in potential efficiency gains in compliance, administration, marketing and customer service. Adopting AI in some form may be needed as a defensive measure to stay competitive on operational costs.<\/p>\n
One example concerns \u201calternative data,\u201d or data beyond that from corporate financial reports and regulatory filings \u2014 including data collected from social media, retail card transactions and smart devices.<\/p>\n
\u201cAdvancements in AI algorithms, coupled with lower computational and data storage costs, now enable the conversion of alternative data into interpretable signals for investors,\u201d Moody\u2019s said.<\/p>\n
The advantages to asset managers from mining this data probably won\u2019t last, but will become table stakes. \u201cThe growing popularity of alternative data will gradually erode its value, compelling asset managers to constantly seek new, exclusive data sources,\u201d the report said. \u201c[T]oday\u2019s alternative data may become tomorrow\u2019s mainstream information.\u201d<\/p>\n
This growing use of alternative data, coupled with efficiency gains, could allow firms to invest in certain asset classes more economically \u2014 private credit, for example.<\/p>\n
\u201cAlthough this segment has grown rapidly, rising to US$1.5 trillion in 2023 from US$1 trillion in 2018, its opacity hinders faster expansion,\u201d the Moody\u2019s report said.<\/p>\n
By making it easier for firms to assess the risks of smaller credit issues, AI could enable more firms to participate in this asset class. \u201cAI could also streamline the analysis of financials and legal documents, which are less standardized than in public markets, and facilitate investment valuation,\u201d Moody\u2019s said.<\/p>\n
Whether any of these developments pays off as handsomely as AI\u2019s proponents hope remains in question. Skeptics argue that the processing power, chips, infrastructure and energy needed to run economically transformative AI models will prove prohibitively expensive.<\/p>\n
A report published by Goldman Sachs & Co. LLC earlier this year concluded that in only the most favourable scenario, \u201cin which AI significantly boosts trend growth and corporate profitability without raising inflation,\u201d would the technology\u2019s adoption drive above-average returns for the S&P 500 in the long run.<\/p>\n
This article appears in the September issue of <\/em>Investment Executive. Subscribe to the print edition<\/a>, read the digital edition<\/a> or read the articles online<\/a>.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":" Gaining an investment edge with large language models is no simple feat <\/p>\n","protected":false},"author":73592,"featured_media":416110,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[3013,3021],"tags":[2629,2403],"yst_prominent_words":[6,14,4147,5033,6665,6725,9143,11191,20491,35179],"acf":[],"_links":{"self":[{"href":"https:\/\/www.investmentexecutive.com\/wp-json\/wp\/v2\/posts\/492348"}],"collection":[{"href":"https:\/\/www.investmentexecutive.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.investmentexecutive.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.investmentexecutive.com\/wp-json\/wp\/v2\/users\/73592"}],"replies":[{"embeddable":true,"href":"https:\/\/www.investmentexecutive.com\/wp-json\/wp\/v2\/comments?post=492348"}],"version-history":[{"count":5,"href":"https:\/\/www.investmentexecutive.com\/wp-json\/wp\/v2\/posts\/492348\/revisions"}],"predecessor-version":[{"id":492502,"href":"https:\/\/www.investmentexecutive.com\/wp-json\/wp\/v2\/posts\/492348\/revisions\/492502"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.investmentexecutive.com\/wp-json\/wp\/v2\/media\/416110"}],"wp:attachment":[{"href":"https:\/\/www.investmentexecutive.com\/wp-json\/wp\/v2\/media?parent=492348"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.investmentexecutive.com\/wp-json\/wp\/v2\/categories?post=492348"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.investmentexecutive.com\/wp-json\/wp\/v2\/tags?post=492348"},{"taxonomy":"yst_prominent_words","embeddable":true,"href":"https:\/\/www.investmentexecutive.com\/wp-json\/wp\/v2\/yst_prominent_words?post=492348"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}