In the technology press, machine learning and A.I. are relatively universally lauded. Most writers, observers and futurists alike attest to the coming panacea — A.I. can solve (or at the very least mitigate) all our problems in one way or another. There’s unlimited potential, and unless you’re afraid of the sci-fi singularity of a Terminator-esque future, there’s not a ton of downside to bother yourself with. I’m not totally without fault on these counts — those of us who see great potential in the emerging technology are right to be excited about its possible uses while we work to hone the raw elements into usable tech. But, it’s also worth seriously considering the shortcomings of these technologies so that we can better apply them strategically (and effectively!). And one of those weaknesses appears to rearing its head in a very, shall we say, expensive industry — hedge funds trading.
In mathematics, statisticians refer to ‘signal’ and ‘noise’ when analyzing trends and extrapolating forecasts. These terms are shorthand and come from signal processing, in which the ‘signal’ is the desired information (like a message or audio transmission, for instance), and noise is the interference in transmission (like static on the radio).
In statistics, when analyzing a dataset, you want to tune out the noise in order to focus on the signal. The signal is the predictive data that if you can isolate correctly, will lead to best (read: most accurate) predictions. Nate Silver literally wrote a book called ‘The Signal and the Noise‘ about the science of predictions (and his track record is damned impressive).
Anyway, I tell you all that because one of the observed weaknesses in machine learning to date? Separating signal from noise.
When it comes to machine learning, the more the data the better. It uses huge computing power to analyze massive datasets and draws inferences from that data. The more data, the better the trends it can identify.
But what if that data is just all noise?
It turns out artificial intelligence still tries to find a trend even if all the data is noise. And one of the most promising applications for A.I. also happens to be a loooooooooooot of noise — the financial markets.
From Jon Asmundssun at Bloomberg:
One of the potential pitfalls for machine learning strategies is the extremely low signal-to-noise ratio in financial markets, says Marcos López de Prado, who joined AQR Capital Management as head of machine learning in September and is the author of the 2018 book Advances in Financial Machine Learning. “Machine learning algorithms will always identify a pattern, even if there is none,” he says. In other words, the algorithms can view flukes as patterns and hence are likely to identify false strategies.
Basically, markets move constantly and a lot of that movement isn’t always rational or business-related — it’s hugely emotional with wild swings based on the overall ‘feel’ of investors. When computers are blindly pointed at all that data, they can deduce trends that aren’t real; it’s truly random.
From the Bloomberg article again, which demonstrates what that looks like in practice: “The Eurekahedge AI Hedge Fund Index, which tracks the returns of 13 hedge funds that use machine learning, has gained only 7 percent a year for the past five years, while the S&P 500 returned 13 percent annually. This year the Eurekahedge benchmark dropped 5 percent through September.”
A.I.-based hedge funds are underperforming the index-fund ETFs that simply track the overall movement of the market. And it might be because the machine-learning algorithms aren’t yet sophisticated enough to separate signal from noise… which is uniquely difficult in financial markets because there’s just so much damn noise!
A.I. could very well prove a game changer in finance, but we have to better hone its foundational desire to find trends, even when none really exist.