Research into improved prediction models
CACEIS explores new methodologies for understanding and integrating human behaviour into macroeconomic prediction models, and the applications for asset management.
Classic macroeconomic models, used between 1776 and the 1990s assume complete rationality - that humans will always attempt to maximise their utility and organisations will always attempt to maximise their profits. These models tend to be strictly positivist, using deductive approaches that typically rely on single methodologies with large samples. Assuming complete rationality however does not always make sense, as human behaviour is not always rational. We all have biases, motives for our behaviours, we all make mistakes and behave irrationally every now and then.
Classical, rationality-assuming macroeconomic models do not explain well the boom and bust cycles, where the only possible cause of those cycles, if irrational human behaviour is discounted, is exogenous shocks. External shocks are notoriously difficult to integrate into macroeconomic models. Rare, with an extreme impact, and only hindsight-aided predictability, are the three attributes of so-called ‘Black Swan’ events. A Black Swan event like the dot.com bubble bursting had an extreme impact – a rough calculation estimates that the cost amounted to US$1.75 trillion. Black Swan events are also not restricted to one specific sector, but can rather occur in many, including weather, technical, economic, political, internal fraud and technological. However, due to such events’ unpredictability, it is nearly impossible to factor them into useful models.
Nevertheless, the observation that boom and busts occur with some regularity allows us to deduce that actual macroeconomic cycles are the result of human behaviour. Black Swan events, on the other hand, are non-normally distributed. More recently proposed models offer an explanation based on a behavioural macroeconomic model, in which agents are assumed to have limited cognitive abilities and thus develop different beliefs. Such models produce waves of optimism and pessimism in an endogenous way and therefore provide a better explanation of the movements.
Recently, central banks and financial institutions, in an attempt to reduce risk and volatility of boom and bust cycles, have started using models that are more flexible in the assumptions on behaviour and policy. For example the OECD, ECB and BoE are using software that allows for movements between forward-looking, rational explanations and adaptive learning for consumers, firms and labour and financial markets. These models have the advantages of allowing for stochastic shocks which means different scenarios can be analysed based on the effects of a given shock on factors such as trade and FDI.
However, in order to re-evaluate our prediction methods, we should look to the advances being made in behavioural economics along with the spread of social media and the internet and how these can be leveraged to improve prediction models. From a purely statistical viewpoint, social media analytics models are more robust than those based on surveys as the samples are bigger and people are less exposed to the bias issue. In other words, behaviours are not influenced by the data collection process. For example, there are more than 200 million Facebook users in the United States (as at January 2017), which roughly represents half of the total population. No survey could ever be based on such a sample.
The question is how to integrate such data into macroeconomic models for prediction purposes. Major advances in technology, such as natural language processing can provide an answer, as they have the ability to process vast sets of text data into meaningful information using sentiment analysis techniques. This data can then be incorporated into macroeconomic models and enable prediction accuracy to be significantly improved
What are the benefits that new methodologies used in prediction models can bring to asset managers? The benefits can be broken down into three areas, 1) Investments, 2) Compliance and regulation, and 3) Operations and Clients. Firstly, investor sentiment on social media can be analysed in order to make better decisions and improve product performance, and machine learning can be used to generate trading ideas. CACEIS’s new data analytics service is already incorporating social media data to benefit clients. Secondly, advances in natural language processing could allow us to better define investor suitability under new regulations being introduced under MiFID II. Models will also help asset managers better predict fund performance in the event of another financial crisis, which is required due to European regulations aimed at strengthening investor protection levels. Finally, such models will improve analysis of client data, helping asset managers enhance their client experience and attract and retain new assets. Alongside this, internal machine learning and big data capabilities will increase internal efficiency and reduce costs.
For more information on prediction models and data analytics, tune in to the “Risk prediction in unpredictable times” session at #FundForum, with a speech by notable pollster, Jean-Pierre Kloppers and an expert panel discussion moderated by Joe Saliba, Deputy CEO of CACEIS.