Humans have an essential part to play, even in the cutting-edge world of systematic investing.
Winton’s investment strategies are best described as systematic, due to the way we use algorithms to determine our positions in financial markets. At first glance, systematic investing can seem mysterious and complicated, especially when compared to investing with human fund managers, who often use stories and anecdotes to explain investment decisions.
In reality, the opposite is true: algorithms are just rules for processing data, which respond to new information in a consistent, repeatable manner. The precision of these rules allows us to judge what portfolios our algorithms would have chosen at any given time during the history of the financial markets – and why.
Imagine how hard it would be for human fund managers to do this – to accurately discern their anticipation of and reactions to events more than 50 years before they were born. A range of doubtful assumptions would be needed in each case to even hazard a guess.
The elegant simplicity of an algorithm is that it removes the need for such heroic supposition, allowing us to run back-tests that help to distinguish a long-term edge from a lucky break. We can also run stress-test simulations of past crises on our portfolios.
This description of what it means to be systematic makes it sound as if all investment decisions could be taken by computers, in a completely automated fashion. You might wonder why we need humans around at all, if, in the age of Big Data, the amount of raw material that an algorithm can work with far surpasses what could be assimilated by a Homo sapiens counterpart.
But systematic does not and should not mean fully automated. Despite the immense power of modern computing, it is neither advisable – nor even possible – to dispense with humans entirely. To take one example: how do we design and choose our algorithms? The clue is in the question. Recent media hype has implied that developments in machine learning and artificial intelligence might allow a computer to select the best investment systems for us. But even if that were the case, it would only push the need for human involvement one level back. After all, there are a multitude of ways to utilise machine learning methods – many different algorithms for choosing algorithms.
There are, of course, tasks that can be automated effectively. Consider, for instance, the way we calculate the suite of risk statistics across all the asset classes that make up our portfolios. For such large scale, recurring calculations it is necessary to harness the computational power of machines. But nuanced interpretation of these risk statistics is still best done by people.
Ultimately, humans have to take responsibility for the decisions enacted by the machines. At Winton, this responsibility is discharged by an experienced senior management team, which approves or rejects proposals for innovative algorithms from our strategy managers and researchers. This structure maintains accountability, while helping to remove the biases and emotions that accompany decisions taken by individuals.
Former chess grandmaster Gary Kasparov's book Deep Thinking, which charts the story of his loss to IBM's Deep Blue computer programme, points out that even today a decent player with a standard computer can beat the best "chess-specific supercomputer". Likewise, rather than giving machines autonomy, Winton's research and investment teams collaborate with computers at every stage of the investment process.
Even data – the basic building block for our investment systems – must be thoroughly quality-assured. No matter how smart the algorithm, garbage in will result in garbage out. Computers are well suited to the initial phase of data-testing, given their ability to apply many checks at speed. When they flag problems, however, the human brain is still better able to cross-reference those irregularities against other data sources, which may not be directly piped into an investment system. This combination of manual and computer testing produces higher quality data that is sourced more efficiently than either human or machine could achieve in isolation.
As these examples show, the notion that human involvement in investment management should, or even could, be fully automated is wide of the mark. For certain specific aspects of the investment management problem, computers excel, and the remarkable advances in computing over the last 30 years have enabled systematic investment firms like Winton to thrive. Yet we continue to believe that for investment it is the combination of humans and machines that offers the most powerful approach.