Advisor Perspectives welcomes guest contributions. The views presented here do not necessarily represent those of Advisor Perspectives.
Artificial intelligence has revolutionized financial engineering. Some hear the hype and expect miracles, but the situation is complex.
Pioneering firms like Renaissance Technologies provide a reality check. The New York– based company gained an early edge in the 1980s with systematic, computer-driven trading. But success did not come overnight.
Renaissance has refined its strategies continuously. Instead of chasing size like many of its rivals, the firm has kept things tight and focused on the basics. It recruits top talent, collects exclusive data, and updates its technology to stay sharp.
Machine learning came later, after Renaissance already had a solid foundation. The result is a competitive edge that is difficult to replicate, one that other large firms are now attempting to build using their own resources. Prominent hedge fund managers like Citadel and Millennium use machine learning for high-frequency trading and other quantitative strategies.
Machine learning also supports mid-frequency strategies by combining signals from different sources, identifying complex links among factors, and making portfolios more robust.
Nobody doubts the potential of machine learning. For portfolio managers, the question is not whether it works in theory; they know it does. The question is whether the expense fits their business size and goals.
Implementation costs are high, mostly fixed, and require significant upfront investment. Before firms can cash in, they need powerful cloud computing, fast processors, and robust data movement. On top of that, they must hire the right talent and pay for exclusive data.
All these costs can drain resources before anyone sees a payoff. Firms must proceed with caution, especially when they have less than $1 billion in revenue. Five mistakes are common.
Mistake 1: Jumping Too Fast
The first misstep occurs when portfolio managers race to machine learning before building a proper foundation. Firms do not need cutting-edge technology to do this. They can use principles established decades ago.
Nobel laureates Eugene Fama and Kenneth French published groundbreaking research in 1993 that still guides quantitative finance today. Additional breakthroughs followed.
Having large teams and expensive tools can make things easier, but systematic alpha typically starts with profitable factor strategies. Strong operations and risk management come next. Machine learning comes third. Portfolio managers should add the technology only when the benefits outweigh the costs.
Mistake 2: Setting Unrealistic Expectations
Once factor strategies are in place and returns are limited by capacity, machine learning can help make useful improvements. But portfolio managers delude themselves if they expect explosive growth.
Even advanced quantitative managers acknowledge that machine learning generally yields incremental rather than transformational improvements. A McKinsey & Company report backs this up, showing that profit margins tend to increase only slightly.
Recent research by Shihao Gu, Bryan Kelly, and Dacheng Xiu reaches a similar conclusion. They show that machine learning can improve out-of-sample return predictions, but these gains are usually small, raising Sharpe ratios by less than 0.2%.
Firms banking on machine learning miracles can go bust.
Mistake 3: Misreading Results
Once output starts to come, portfolio managers must make sense of the data. This is a separate challenge.
The main practical uses of machine learning are in modeling phenomena that traditional factor models do not handle well, such as changing factor exposures over time and forecasting risk. Models like these only work if they are understandable. Machine learning models that find factors without economic meaning are likely to fail when tested on new data. This is why firms like Renaissance invest in talent.
Getting bad advice is easy. Finding experts who can correctly interpret the output is hard.
Mistake 4: Amplifying Errors
While AI is great at spotting patterns, this strength can also be a weakness. If firms feed messy or biased data into the system, problems are amplified.
For example, analysts can leave failed companies out of historical records — a pitfall known as survivorship bias. They might also fall victim to look-ahead bias by using information that was not actually available at the time, or overlooking subtle shifts in how companies report their financials.
Machine learning models excel at finding all kinds of trends, even the ones that are not real. Unlike simpler approaches, advanced models can make predictions that seem confident but miss the mark if the data are not clean.
Avoiding blunders takes more work and money than most people realize. Firms must clean their data, check for biases, make sure they are looking at the right snapshots in time, and keep everything up to date.
Many managers get caught up in building fancy models and forget that basic data quality is just as important. In fact, splitting resources evenly between keeping data clean and working on models usually leads to better long-term results.
Mistake 5: Overlooking Alpha Decay
Classical asset pricing theory holds that, as more money flows into profitable strategies, returns shrink due to limits on temporary price differences. Put simply, investors adjust, especially when new research appears in academic journals or on industry forums.
The result is alpha decay, the decline in a strategy’s profitability as it spreads and becomes familiar. Alpha capacity becomes the primary concern. Large multimanager hedge funds can impose competitive advantages through superior infrastructure, execution capabilities, and liquidity access that smaller managers cannot match.
This concentration dynamic further compresses available alpha for the wider systematic investment community.
Systematic asset managers can respond in five ways when considering machine learning and other AI strategies: build strong factor strategies grounded in empirical asset-pricing research, set realistic expectations, hire smart people, invest in data quality, and adjust for alpha decay.
Artificial intelligence has pushed asset management forward, but it is far from a magic fix.
Iván Blanco, Ph.D., is an associate professor of finance at CUNEF Universidad Madrid, an instructor at WorldQuant University, and founder and director at Noax Capital.
A message from Advisor Perspectives and VettaFi: Discover something new! Click here to register for our upcoming webcasts.
Read more articles by Iván Blanco, Ph.D.
Advisor Perspectives welcomes guest contributions. The views presented here do not necessarily represent those of Advisor Perspectives.
