The latest trend in systematic investing is machine learning. This resulted from the fact that technological progress made very high computing power and almost unlimited data and storage capacities available. According to Harvey, open source software also played an important role: "This made software developers much more efficient because they didn't have to reinvent the wheel every time. They can now easily access GitHub, where there are freely available solutions as well as a community that deals with similar issues.
Clear advantages...
Harvey cites the discipline involved as the biggest advantage of systematic trading. Algorithms are based on rules that they implement emotionlessly and without distorting behavioural effects. For example, they do not unnecessarily hold on to losing trades, which human traders do (disposition effect). Even in phases of strong turbulence, algos always keep a "cool head" and can even profit from mistakes or behavioural effects of discretionary traders. This is because emotions such as euphoria or panic show themselves in certain patterns and can thus be systematically exploited accordingly.
Algorithms are based on rules that they implement emotionlessly and without distorting behavioural effects.
Another advantage of machines is their enormous capacity to process information. Given the explosion of Big Data, discretionary managers today can no longer process the volumes of data without computer support. In addition, algorithms do their work extremely quickly and relentlessly efficiently. They are therefore also able to immediately recognise short-term effects such as important news in the market.
... but also great challenges
Algos are not perfect, however. They represent a simplification of our complex world and are often highly parameterised. By design, they have been adapted to a certain degree to past data, which limits their flexibility. Due to the poor signal-to-noise ratio in the markets, many strategies optimise - often unconsciously - the random noise instead of the true signal. This results in over-optimised algorithms that look good in backtesting but disappoint in live trading.
Added to this is the non-stationarity of the markets. The world changes, as a well-known saying in the markets goes: "This time is always different". Optimising a stationary algorithm to a non-stationary market is therefore doomed to failure. Alternatively, however, even with today's technology, constructing a reliable algorithm that adapts appropriately over time to a changing market environment is an enormous challenge.
"This time is always different.”
Another challenge described in the paper is the black box problem. This often occurs with machine learning applications that are purely data-driven instead of being based on solid economic correlations. Investors should therefore be wary of statements by managers who cannot or will not reveal how their model works. After all, even the most complex algorithms can be traced to some degree - for example, by feeding shocks into the input parameters and then analysing the change in outputs. Harvey concludes that all algorithms should therefore be explainable.
Last but not least, obtaining and cleaning large amounts of data involves considerable effort and high costs. One example is the necessary IT resources for reliable database management as well as sufficiently high computing capacities. The quality of the processed data also requires the utmost care. Even the best algorithm can only achieve good results if its database is correct.
You can't do it without experts
Even if all the pitfalls and challenges are taken into account, there is one thing that algo trading cannot do without: human expertise. For as Jack Forehand writes in his article "When Quantitative Becomes Discretionary", there is a big difference between merely reducing human decision-making in the investment process and eliminating it altogether. [2] While many a model seems to excel when compared to discretionary strategies, the truth is that even systematic strategies are still shaped by discretionary influences. This is because quantitative models cannot simply run on autopilot.
Quant models cannot simply run on autopilot.
According to Forehand, the more a strategy includes concrete rules - or sets them up itself to then determine its actions - and the more consistently these rules are followed, the higher the probability of attractive long-term returns. But in the end, there is always a person or a whole team behind the rules or their creation basis. And it is at this level that a number of decisions about strategy have to be made, which have a significant impact on how the whole thing develops over time.
He describes three examples of how human decision-making plays a role in even the best quantitative strategies:
Initial design: this is where we determine what goes into the strategy and how it is defined. What metrics and combinations are used and how are they to be weighted? What is the investment universe, how many positions are to be held under what conditions, and when or how are rebalancings to be made? These and many other complex decisions require a well thought-out decision-making process - and thus human expertise in system development.
Ongoing development: Even successful quantitative strategies cannot be operated permanently without adjustments. This is mainly because real-world market conditions change, requiring the flexibility to continuously update strategies. In the vast majority of cases, the decision on whether there is appropriate evidence in the data to make an adjustment still has to be made by experienced practitioners.
Pulling the rip cord: Most quantitative strategies can remain profitable over the long term through regular reviews and necessary adjustments. But it can also happen that the foundation is based on something that no longer works in principle. Human experts are indispensable for making informed decisions about when to pull the ripcord.
The future of system trading
According to Campbell Harvey, systemic investing will be here to stay and will continue to grow in importance. However, one should not underestimate the complexity required for the successful implementation of algorithms. This results from the necessary interaction of a distinctive competence in the classical investment area, the required expertise in the area of quantitative data analysis as well as the necessary technical equipment.
Human experts remain of central importance, especially in system development. This also applies to seemingly self-sufficient machine learning, for example, in order to initially select the most suitable approach for the respective strategy from hundreds of available models. Furthermore, the models must first be successfully trained and later executed in real time.
Human experts remain of central importance in system development.
So the idea of a purely quantitative strategy is more of a myth than a reality. It is therefore just as important to evaluate the people and process behind a trading system as the strategy itself.
Conclusion
Quantitative models are a great asset in the investment process. Algorithms help implement strategies consistently and resist the siren song of emotional decisions. However, this does not mean that everything can or should be quantified. It is more about finding the right level of discretion within the overall investment process.
Ultimately, even the algo world cannot function without people. Technology alone does not increase the likelihood of outperformance in the market. This ultimately depends on the skills of the team applying that technology. So despite all the advances in systems trading, we are miles away from completely eliminating human decision-making from the investment process.
