Alternative Investment Management Association Representing the global hedge fund industry
Sabre Fund Management
What is quantitative or “quant” investing? In short, it is where mathematical models rather than manager’s discretion make investment decisions.
What do quant managers do apart from running sophisticated computer programs? An analogy with cars may help here. No matter how sophisticated a car, it still needs a driver and sometimes a mechanic. The driver must know when to brake and when to accelerate. Quant managers provide both functions – they control risks as well as fix things such as data errors when they occur. Most importantly, every car needs to be designed and engineered in the first place – the most important part of a quant manager’s job.
There are two types of quant managers: ones that aim to produce performance in line with the market and ones that aim to beat the market. The former are well established because they can replicate indices without having to own each stock, thus making savings in transaction costs. Some argue that is what all quants should do but we do not share this view. This article aims to highlight the advantages that quants have in active management over their discretionary peers.
Advantages of quant strategies
A quant investment process relies on a scientific approach, with strategies researched and tested before implementation. Even then, only empirically proven strategies are given a chance in the real world. The longer the testing period, the better, as empirical evidence becomes more reliable. Quantitative investment processes are often perceived as lacking human judgement. On the contrary, human judgement plays a critical role during the model building stage. The main difference between a quant and a discretionary manager is in the timing and not the presence of the judgmental input. Intuition, experience and understanding are important in research but not a feature in implementation.
The implementation is model driven and therefore emotionless – we know that emotions and trading do not mix well. In fact, emotions often cause behavioural price anomalies (such as overreaction to recent information) that can be exploited by quant models. At the same time, quant models can trade more cheaply and efficiently. Indeed, program and algorithmic trading procedures are already recognised as adding value compared to a discretionary approach, with their penetration now widespread amongst quants and non-quants alike.
Improvements in computing power mean that better timeliness can be achieved through analysing new information more quickly for thousands of stocks. In contrast, discretionary managers can hold a rather more limited number of “ideas” at any one time.
More accurate volatility targeting is provided by quants that exploit a linear relationship between leverage and volatility. Discretionary managers typically seek higher volatility through concentration, but this more complex relationship (i.e. quadratic) makes volatility targeting much more difficult. Thus quants yield better returns relative to the risk they are taking. This also implies better and easier structuring of leverage and portable alpha products for institutional clients.
Risk testing and control is also easier and more accurate in a simulation-based environment and few quant processes do not actively manage risk. Discretionary managers will admit to using quant methods for risk measurement. However risk measurement is one thing and risk management another – merely measuring the risks might not lead to any action to mitigate excessive and/or unfamiliar risk taking. Whilst it is true that measures such as the Sharpe ratio have their limitations, many discretionary managers may adopt a rather dangerous approach: “If we look after the return, then the risk looks after itself”.
Scalability and assessment of capacity is also readily achievable in a quant framework. In addition, a repeatable systematic approach should lead to more consistency and less style drift – a constant problem in successful hedge fund investing.
For all these reasons, one does not have to be a quant manager to understand the advantages of quant investing. However, quant investing is not without its challenges and the next section discusses some commonly stated potential weaknesses.
Potential drawbacks of quant strategies
“Quant is difficult to understand”. We hope that the current generation of quant managers will point first at similarities and then at improvements to the traditional discretionary approach. We further hope that this article addresses some of these points.
“Data is dirty”. So, “quant models are no more than garbage in and garbage out”. Most of the data has been provided by respected data providers, gathered over many years at substantial costs. Armies of analysts are employed to ensure data cleanliness and accuracy. Furthermore, modern statistics are well equipped to highlight and correct anomalies so that garbage is filtered out before it is input to the models. The data is also validated by the success of the models that use it.
“Quant is Data Mining. Models work well in sample but they are poor out of sample”. We have already pointed out that models should be based on sound theory and backed by empirical evidence in terms of out of sample testing. Look-ahead and survivorship biases should be avoided by simulating decision making purely based on information available at the time of the test.
Optimisers are numerical problem solving engines used by quants for asset allocation and/or stock selection. Typically, there are two types of criticism. The first is ignorance based: “We don’t understand this, so why bother? ”, equivalent to saying: “Why drive a car when a horse would do?” The second type is informed, as we cannot deny that some early generation optimisers are “error maximisers ” – they have substitution problems, fail to account for non-normality in returns and cannot cope with regime changes. However, through many years of academic and practitioners’ research, enhanced optimisation techniques have been developed to solve these issues. We now have optimising tools that can build robust portfolios, with success confirmed by the performance of funds that use them.
“Sample size too small”. In general, the data set for testing should be as extensive as possible and should include as a minimum a full economic cycle and various market conditions. Some early quant funds failed because they limited testing to too short a period. Fund allocators should also consider that rushing to invest in funds with little or no history could lead to similar problems.
“Quant models can be copied and as a consequence, their alpha will decay as exploitable opportunities reduce.” Firstly, this acknowledges that models are actually worth copying. Secondly, managers can protect their intellectual property to a large degree. Thirdly, we believe that the additional data more then offsets this potential drawback and to explain this, we devote the whole of the next section to this topic.
Why should quants improve with time?
Rather than allow the existing models to become stale, they are recalibrated to reflect new data as it becomes available. At the same time additional strategies can be added to the process, for example by applying the existing models to new markets. Quants can also include models that rely on data that can only be obtained in real time, as no appropriate historical databases exist for it. For obvious reasons, such proprietary strategies are often called “the Insider Models”. Clearly, as history lengthens these processes will invariably improve. All of this should more than offset any alpha decay. Furthermore, the stronger the investment process, the more it makes sense to increase return and volatility targets.
Another analogy can be made between computers in investing and in chess. Recently, some formidable chess matches were played between the world champion (Kramnik) and the computer (Deep Fritz). At first, the computer was losing, but the penultimate match ended in a draw. Since then, further enhancements were made to Deep Fritz algorithms and to make it fairer, many new ways of using I.T. for analysis were allowed to Kramnik. The last contest took place late last year and it was won by a large margin by Deep Fritz. Despite being the best chess player in the world, Kramnik could not compete with the processing power and knowledge based analytical capability of the machine.
Chess is an old game and its theory has been developing for hundreds of years. All this knowledge is available to Deep Fritz, which in its memory has over a million games. Investing is much younger, with quant investing only a few decades old. There is still much knowledge to be gained and many improvements to be made to the algorithms that drive our investment practices. Combined with ever-increasing computational capabilities, we expect greater dominance of quant investing in the future.
The message is clear. “Past performance is not indicative of future performance. In the case of a good quant, it should get better!” Ignore established and successful quants at your peril. However, behind any computer program there is always a human brain. Investing in a quant approach is the best way to profit from human ingenuity but without the danger of human error.