Traders using machine-learning models were able to recover from declines, but experts warn of weaknesses
The recent market turbulence from the coronavirus caught many investment managers wrong-footed, including those that use artificial intelligence to inform their trades and stock selection.
Citing data from Eurekahedge, the Wall Street Journal reported recently that funds with investment processes underpinned by AI lost money during a marketwide selloff that kicked off on February 19.
Based on an index of 23 such funds tracked by Eurekahedge posted a 2.62% loss in February, compared to the 1.87% loss for the broader Eurekahedge Hedge Fund Index. The S&P 500 fared worse, declining by 8.41% for the month.
But in March, the wider index posted losses of 6.11%, while the AI fund index notched a gain of 2.11%. The U.S. stock market, meanwhile, shed 12.51% as per the S&P 500.
The reversal among AI funds in March, the Journal suggested, implies that the models they rely on were able to adapt to changing market conditions with the incorporation of new data and decreased dependence on historical data.
Many critics remain unimpressed. They contend that machine-learning models used to simulate interactions between stocks, bonds, and derivatives rely on historical data and assume specific market conditions. The market stressors of the past, detractors note, were usually confined to single events, or a series of economically related events with one trigger in common.
Then there’s Mohammad Hassan, the head analyst for hedge-fund research and indexation at Eurekahedge, who noted that companies within the index showed wide dispersion of results, with some experiencing double-digit gains as others swung just as strongly in the other direction.
“The constituents of the index also run the gamut between firms that are essentially quantitative or systematic managers using some AI through to technology-focused funds developing neural networks and deep-learning models,” the Journal said.
While accepting that AI is no magic bullet for investing, firms relying on such strategies contend that its limitations can be mitigated by high-quality data inputs and the use of human intelligence as a check and balance.
“There’s a difference between handing over an AI and its learning process to a freshly minted grad with a degree in computer science and mathematics and saying, ‘Teach this AI,’ versus a stable of market veterans,” said David Aferiat, co-founder and managing partner of Trade Ideas.
