4 questions US institutional investors can no longer afford to ignore

[00:00:04.98] Our team has been using machine learning, a subset of AI, to enhance our investment and decision-making process for several years now. Machines are generally great at identifying patterns, adapting to new information, and processing large amounts of data at scale. And so logically, recent advancements in natural language processing have further improved our ability to interpret and extract insights from large volumes of text data.

[00:00:28.58] So in terms of AI in the investment process, we have found a variety use cases to be particularly interesting.

[00:00:35.19] One is Predictive Analytics, where we apply machine learning models to analyze large datasets to predict future market movements.

[00:00:43.09] So for example, a big driver of future stock prices is future profits and growth. So each day on Wall Street, thousands of analysts provide forecasts on fundamentals like earnings, sales, and updated analyst expectations. So these forecasts are very closely watched by investors, therefore, we track these forecasts across nearly 90 countries very closely, which is a lot of data to sift through.

[00:01:03.62] So using AI, we developed a Natural Language Processing signal that seeks to extract certain features from that data such as patterns in investor sentiment or risk appetite to ultimately offer different, market-movement predictions.

[00:01:16.30] Another application of Large Language Models is to identify the dominant themes in markets and from there build customized baskets of stocks that are complementary to that theme such as AI beneficiaries or defense stocks.

[00:01:28.36] A second use case we focus on is Data Validation. “Garbage in, garbage out” we like to call it. We invest a lot of time, money, talent, and resources to develop our signals, all of which are subject to rigorous testing, review, and team debate before being put into production on behalf of our client portfolios. Data integrity is crucial. If the data that trains and feeds the models is flawed or incomplete -- the output can be compromised. AI assists in data validation by identifying and adjusting anomalies in our input data, which is crucial for the accuracy of our quantitative models.

[00:02:08.12] While machines are great at consuming large amounts of data, they don't understand it. Even the most sophisticated Generative AI models just don't “get” it. Yet, they are spectacularly good at guessing. So we build proprietary, systematic signals that form the foundation of our decision-making. So these signals encapsulate and generalize our knowledge as seasoned investors and because we build the signals, we're well aware of their strengths and weaknesses and when and how they should be applied. So as sophisticated as AI and broader computing is these days, it still can't reliably or sustainably run a high-conviction portfolio without human supervision and intervention, when needed.

[00:02:43.88] So machines will struggle with predicting and properly estimating the risks of “edge” cases, or unique situations that fall outside its training data. So for example, if you train a

fixed income signal over a historical period where inflation has been benign and interest rates have steadily decreased, it can struggle when it gets new data post the pandemic that is very unlike anything it has seen before. So and as we know, it's usually those types of unprecedented events and dynamic market regimes that can make or break your portfolio and are the most important times to have a human hand on the wheel.

[00:03:19.63] One is Biases in your model. AI models have biases based on the data they've been trained on and the assumptions that were used in the model's construction. Examples of biases on input data may include earnings transcripts painting a rosy picture of the future, or the public speeches of most CEOs that tend to be more optimistic in nature. So, if you apply sentiment analysis to an earnings transcript without accounting for the positive bias, then you may not get the result you expect. Another risk is the Overfitting of models. As investors, we are essentially trying to predict the future. And yet markets are often driven by short term psychology, external events, and random fluctuations. As such, all of this 'noise' may mask the more meaningful information that a quant process seeks to capture. But the model doesn't know this. It just looks for the pattern in the data even if the data is dominated by noise. So the result of this could be an overfitted model that looks strong when applied to historical data but fails to deliver robust performance when fed with new data or a new regime that it hasn't seen before. And so it's important to understand the implicit assumptions being made and the sensibility beneath the data and not just get wrapped up in a historical back-test.

[00:04:28.58] And lastly, the risk of Transparency or interpretability. So as helpful as models and cleverly designed quantitative signals can be, there can also be a bit of a “black box” element on how exactly some of the models come to their conclusion. It can be hard to build trust and confidence in an output if you don't have visibility into exactly how it was generated because of the models' complexity. So in the case of LLMs, 'hallucinations' are also a well-known risk, which is when the model, very confidently, produces very plausible-sounding nonsense.

[00:04:58.97] And so to close out, as investors, we need to account for these types of limitations when calibrating and using these types of signals. A balanced approach is crucial here, where AI can serve as a complement to human judgment, but never as a substitute.

[00:00:05.15] Since the late 1980s, the era of hyper globalization has led to increased, global coordination across economies and a lowering of economic inequality across developed, and developing nations. However, this era has started to revert post-2020. Rising protectionism is reducing global trade and growth, creating risks to global supply chains and raising costs and inefficiencies. As such, it can also elevate policy uncertainty. And so, altogether, these global shifts should broadly increase volatility, reduce global growth, and intrench inflationary pressures.

[00:00:43.81] Well, how nations respond to the new era will likely lead to differentiated opportunities, creating investment dispersion. For example, more local investment in sectors of strategic priorities such as defense and technology in the United States is likely to create tailwinds for those areas, especially in periods of heightened geopolitical risks. And, from a global perspective, being more selective around country exposures is likely to become more important as well. For example, countries that are more focused on maintaining strong trade linkages across factions are likely to benefit from higher economic growth.

[00:01:22.42] Understanding the interlinkage between evolving geopolitical fragmentation and macroeconomic drivers is likely to present opportunities. If dispersion increases, redefining portfolio allocations across more granular exposures, as well as being nimble, should add significant value, especially in a world where volatility may start to normalize from the ultra-low levels of the prior decade.

[00:00:05.39] For investors who choose to, there are so many ways to facilitate or financially benefit from transition investing.

[00:00:13.81] One of the key benefits of a multi-asset approach is our ability to look across multiple trends.

[00:00:20.23] So for example, AI is a big trend that everyone is talking about. And actually, AI does tie into the transition to a low-carbon economy.

[00:00:29.52] There was recently an IEA report which said the electricity required for a ChatGPT request is 10 times that of a Google request.

[00:00:38.87] So in other words, as AI becomes more and more integrated into our everyday life, we can expect a massive reallocation of capital as energy systems are re-wired and infrastructure needs are met.

[00:00:50.95] And this is even more true when you think that electricity demand may also increase if, for example, more people choose to buy electric vehicles or hydrogen production increases.

[00:01:01.83] The fascinating part about the transition is that there are multiple ways to invest in it, from the new technologies that are scaling up, the inputs that needed to build it, to the infrastructure required to support it. As such, we have the expertise and capabilities to invest across early-stage private markets all the way through to equities and fixed income, depending on the opportunity.

[00:01:24.94] And so from a multi-asset perspective, to bring this full circle, we’re able to look across all of those trends and think through the investable opportunities, allowing our investors to access the transition using a diversified and risk aware approach.

[00:01:46.32] Some investors choose to invest thematically when it comes to low-carbon transition opportunities.

[00:01:51.88] For example, in electric vehicles or renewables.

[00:01:56.01] One thing we’ve learnt is that these themes can be sensitive to the broader market.

[00:02:00.60] So for example, if interest rates fluctuate, this can have an impact on short term performance, even if we have a constructive view on an investment.

[00:02:10.08] By taking a multi asset, top-down view across sectors, our portfolio management team aims to manage these risks as we seek to deliver a smoother financial outcome for our clients.