Drive alpha (α) in the financial sector from deep learning
Artificial intelligence seems to be quite the rave these days. From self-driving cars to smart toasters, there is a strong drive in the world of technology to make computers and machines more human-like in reasoning, but without the errors common to humans. The financial sector like many others is a beneficiary to this innovation.
The practical application of artificial intelligence in investment has the potential to completely revolutionise the industry, creating a sort of distinction between investors who make use of artificial intelligence and those who prefer the traditional approach, with the former having a noticeable edge over the latter.
The whole ambit of artificial intelligence is the development of systems which rival and in some cases surpass human beings in reasoning and intelligent behaviour, coupled with a penchant for learning. This is a significant improvement on machine learning which simply focuses on developing machines that can deliver output based on existing data without the need to specifically program them to do so.
Fundamental AI concepts
Two underlying principles of artificial intelligence are Learning and Integrated Reasoning. Just like humans after which much of it is modelled after, artificial intelligence has to be able to make use of its past experiences to inform the decision it is being faced with. Thus, the machine must ‘learn’ from data gathered from previous functions, current occurrences in the world and the general environment it is exposed to and apply these learnt details intelligently within the bounds of the guiding principles set by its user to the task at hand.
Furthermore, AI must also inculcate within its processes – Adaptation and Evolution. This implies that AI should be able to seamlessly adapt learnt expert investment models to its future investment decisions. AI must also evolve in that it should get better at making sound investment decisions as it continues to learn and interact with investment-related data.
These above-stated concepts substantially set AI apart from traditional quantitative strategies (traditional quants). Traditional quant investment is by design, factor driven and makes use of a simplified, linear approach to investment. AI on the other hand, is able to pull information from a multiplicity of sources and integrate same into investment decisions by identifying the common factors or collective behaviour of these different models, and selecting those that are most relevant to its decision making. In other words, while traditional quants make guesses (educated guesses but guesses nonetheless) due to their limited sources, AI is able to foreshadow and make projections of the possible future occurrences in investment from the pool of sources it draws conclusions from.
The AI edge as opposed to machine learning and traditional quant
Artificial Intelligence fuses the best of computing with the best of human intelligence while retaining little to none of the flaws. As such, compared to traditional quant strategies, AI is more fluid and dynamic in its processes and application, as it appropriately employs data-driven models (as with machine learning) coupled with guiding principles known and/or similar to that of humans.
Although machine learning is an improvement on traditional quants, it suffers its flaw from over-reliance on input data. It may not be able to cope with a fast changing investment system in the long run due to the fact that its data sets will get outdated from time to time and would need to be updated. Artificial Intelligence on the other hand simply needs to abide by its guiding principles to formulate working strategies using the information it has previously learnt. By so doing, it evolves and adapts to various market conditions as time goes by.
The aim of any machine learning model is to be able to generalise its training data set to problem data. This gives it the ability to make predictions subsequently on data it has never seen. Two banes to this generalization of data and also the biggest causes of poor performance of machine learning are ‘overfitting’ and ‘underfitting’. Overfitting occurs where a machine learning algorithm models the training data too well, i.e. it learns the detail and the noise in the training data to the point that it negatively affects its performance on new data. Underfitting on the other hand occurs where the machine learning algorithm is neither able to model the training data nor generalize to new data. Artificial intelligence however employs the use of fundamental principles or rationales to guide its data-driven processes which acts as a failsafe against overfitting or underfitting a particular data set.
Artificial intelligence with its combination of data-driven methods and guiding rationales is able to generate precise, stable and accurate returns. Therefore, in most situations, AI will outperform both machine learning and traditional quant strategies in generating a stable high alpha at lower turnovers.
The hype surrounding the use of artificial intelligence in investment is well deserved as can be seen above. It is able to consistently deliver top-notch investment results using human-like rationale, while avoiding human pitfalls and errors in judgement. This very factor makes AI a possible replacement for human investment managers.
Furthermore, considering the possible applications and benefits of artificial intelligence, coupled with the fact that it can easily outperform other contemporary systems, it is quite tempting to adopt AI methods completely thereby rendering redundant human investment managers. Unsurprisingly enough this is a common concern among sceptics of AI.
Truly AI is able to consistently generate a stable high alpha with lower turnovers, this however does not eliminate the need for human investment supervisors. This is more applicable in an AI’s early stages where it is still ‘learning’ its environment. At this point, a human-machine interaction is essential to check and confirm that the outcomes of the AI’s processes are as desired. However, with long term AI investment strategies, the need for human intervention gradually decreases as time goes on and as the AI further adapts and evolves to fit its role.