Finance is not the same is it was a few years ago. Whereas the Wolf of Wall Street epitomised the traits of the successful trader in the twentieth century, today things look very different. Computer scientists in sweatpants – quants – design and implement complex mathematical and statistical models that more efficiently price securities, reduce risk and generate profits.
One of these relatively new statistical applications is machine learning. A 2020 paper in the Review of Financial Studies by Shihao Gu, Bryan Kelly and Dacheng Xiu use the Sharpe ratio, a measure of the risk-adjusted return of a financial portfolio, to show the benefits of machine learning: ‘A portfolio strategy that times the S&P 500 with neural network forecasts enjoys an annualized out-of-sample Sharpe ratio of 0.77 versus the 0.51 Sharpe ratio of a buy-and-hold investor.’ Machine learning, they find, more than doubles the performance of a leading regression-based strategy from the literature. That’s huge!
But is this just an academic artifact – or does it actually have real-world applications? Hudson and Thames is a London-based engineering company that builds machine learning algorithms for financial investors. I asked its South African founder and CEO Jacques Joubert about the benefits of machine learning: ‘Machine learning’s main use is its ability to model non-linear processes. Linear regression remains the primary workhorse for financial modeling. As one of the leading financial machine learning experts, Marcos Lopez de Prado, notes, we live in a strange world where one half of finance believes that markets are efficient and passive investing is the key. The other half are active managers, the majority of which believes that a model as simple as linear regressions, is capable of harvesting billions worth of dollars of alpha. But it is unrealistic to assume that something as complex as financial markets follows a linear process. This is where machine learning can add value.’
Is that where Hudson and Thames come in? ‘We are mainly focused on implementing the research of others. We often comb through the academic literature looking for breakthroughs or applications of important algorithms and then reach out to the original authors and build out the tools with their guidance and support. Since we have such a large collection of algorithms, we are often able to combine them in new ways to unlock additional capability.’
I ask Joubert about a nagging critique against quants: While it is much easier to predict the future in a period of relatively stability, what happens during a crisis? Are there not greater risks when low-probability events occur when the models are fine-tuned on historical data that do not include such events?
Says Joubert: ‘It is important to remember that every trader and active fund manager uses historical data to forecast the future. Quants fit models to the data – and sometimes, because of the non-stationary nature of financial markets, there will be a structural break. But its usefulness really depends on the question being asked. As humans, we always look for heuristics. There are quant funds that have done very well in the crisis. It is perhaps too sweeping a statement to say that Covid-19 has changed the market forever: for which strategies, which anomalies, which asset classes and sectors? It is not true for all.’
The more I talk with Joubert, the more I realise that his work is firmly rooted in the scientific method, a departure from the ‘finance as an artform’-approach that the old school investors often propagate. I ask him about the synergies between academe and industry. ‘Some people believe that academia lags industry and I can tell you from first-hand experience that this is only true for the world’s best funds like Citadel, DE Shaw, 2 Sigma, and Renaissance. However, the vast majority of the industry has failed to adopt algorithms and can benefit greatly by reading more of the academic literature. Information is slow to disseminate and I believe that a large advantage can be gained by tracking the literature.’
This brings me to a touchy subject. The finance industry has been criticised for pulling in the best talent in what is arguably a zero-sum exercise: when your team wins, somewhere else another team loses. I ask Joubert to make a case for finance – and the advice he would give a student.
‘This is a very interesting question and one that I have given considerable thought to myself. Finance does attract a lot of talent that was destined for other domains. And there is good reason for this: it is largely because of the attractive packages offered. But I do think that finance is a field that gets a lot of unfair criticism. People often neglect to recognize the massive importance that financial systems play in our society. If you own a house, have a student loan, invest in a pension plan, or have raised money for your business via an IPO it is the finance industry that you have to thank.’
And, I want to add, a bad mistake can destroy millions of livelihoods in an instant. You want smart people managing your savings.
‘My advice for students is to think long and hard about your career. Most people give almost no thought to what they will study and simply select a university or faculty for prestige or pressure from their parents. And after you graduate, don’t be scared to gain international experience in a major city such as London, New York, San Francisco, Hong Kong or Singapore. By building a career abroad you will dramatically increase your chances to learn from the world’s best talent.’
I cannot help but conclude with a question about the future? What can his machine learning tools tell us about what to expect in 2021?
‘I wish I had the answers. The tools that we work with do not require us to make forecasts so far out. My personal recommendation is that enthusiasts steer away from using machine learning to forecast the next day’s returns. I have yet to see or know anyone to get that right, asset pricing is an unsolved problem. Where I have seen ML add value is in portfolio management, lowering market impact, transaction costs, forecasting the limit order book, identifying viable trading pairs, and building long-short portfolios.’
In fact, this is exactly the conclusion Gu, Kelly and Xiu also reach: ‘The overall success of machine learning algorithms for return prediction brings promise for both economic modeling and for practical aspects of portfolio choice.’ Machine learning in finance is here to stay. But as ever, it will only be useful if we ask it the right questions.