Despite their prevalence, human forecasters are largely unreliable. This point was confirmed by political scientist Philip E. Tetlock, who examined the quality of 28,000 predictions made over a 20-year period by 284 experts, whose number included professors, journalists and government officials with various political preferences. He found their forecasts were barely more accurate than chance, and worse than basic computer algorithms.
Many industries have known this for years. It is hard to imagine an aircraft manufacturer relying solely on expert opinion or theory when designing a new plane. Instead, it would use a host of experimental approaches: ground-based structural pressure simulations, wing flex tests and cold and heat trials to identify and resolve weaknesses after analysing data obtained from computer modelling.
Winton has been at the forefront of systematic asset management since 1997. Our business is grounded in the belief that profitable insights about financial market investing can be derived and refined through empirical observation and rigorous testing - akin to the techniques employed by the aerospace industry and others.
These techniques can be applied through an approach known as the scientific method. The scientific method involves gathering large amounts of data – in our case about the world’s markets, economies and companies – formulating hypotheses and then rigorously testing those hypotheses using the data collected. A hypothesis is an idea – for example that cold weather predicts higher energy prices – that is framed in such a clear and unambiguous way that we can prove it to be wrong; that is, it is subject to being falsified.
To identify original and potentially useful hypotheses, it helps to view the world from many different perspectives. Winton has a large research and development team, but we employ few economists or financial analysts. Instead, our research and data scientists are drawn from fields such as robotics, particle physics, aviation, statistics, electrical engineering and history.
It also takes experience to get the most out of the scientific method. For one thing, failing to clearly define hypotheses in advance leaves the data liable to be tortured until it provides false confessions. For another, checking a hypothesis against too little data can sometimes throw up false positives. Ideally, therefore, we look to test our ideas back over many decades. The more historical data that our ideas prove themselves against, the more confident we can be of their predictive power.
To that end, we collect data about global equity markets, bond markets, currencies and commodities. In all, our databases contain around 600 million prices across a broad range of financial instruments. The historical price charts displayed in the Winton Chartroom in our London Headquarters stretch back as far as the 1800s for many major markets.
Over the 20 years since Winton was founded, the world has moved our way: more and more data is available for us to feed into our machines in an attempt to better forecast markets. One source is so-called “alternative” data - that which is produced, for example, by satellites or sensors on agricultural equipment. Winton has been using this type of data for some time - for example, using satellite data as an input to a crop forecasting system - but the bedrock of quantitative investing continues to be data coming directly from financial markets.
An unglamorous and often overlooked aspect of quantitative investing is ensuring that the data used to test hypotheses is not just big, but error-free. For example, Winton buys data on the share prices of S&P500 companies from six different suppliers, and we regularly find inconsistencies between those vendors. This is why we run thousands of tests on our data every day before it is used in our systems.
Winton’s rigorous collection, scrubbing and organisation of data underpin our application of the scientific method. Our two decades’ experience of employing this systematic approach is what lies at the heart of Winton’s proposition.