Category: Evolution in employee benefits and advice
Describe your evolution:
In essence, Paul Nixon of Momentum has developed a hyper-personalisation engine powered by personality science and machine learning that feeds the behavioural finance function. Psychology and risk behaviour theory shows that people have long-term risk attitudes (they like it or they don’t), but in the short term, risk feels very different to each and every one of us (exacerbated by cognitive and / or emotional biases). This often leads us to act against these long term risk attitudes and incur a behaviour tax or worse investment outcome when compared to doing nothing at all.
The behavioural finance function is centered around 4 ‘behaviour tracks’ or different investment behaviours and the hyper-personalisation engine will be designed to give the right message to the right person at the right time. I will deal with 2 of the 4 behaviour tracks here.
Behavioural track #1 : Stay invested
In order to manage or eliminate the behaviour tax (from moving money around) we need personality science and unsupervised machine learning to help understand long-term risk attitudes and supervised machine learning to help predict behaviour in-the-moment.
Innovation #1: Money Fingerprint psychometric (reliable and valid) profiler was built over 2 years to assess:
1. Client risk attitudes and risk behaviour
2. Money attitudes (relationship with money based on money scripts or past money behaviour)
3. Personality traits (client default factory settings that relate to investment behaviour such as the degree to present versus future orientation). The marshmallow test showed that some of us are just better (have better factory settings) than others.
How does this help the behavioural track? We can help target clients with low conscientiousness, low money prudence, high money anxiety and high neuroticism with timely nudges to keep them invested. The content is hyper-personalised based on the personality traits.
Innovation #2: Unsupervised machine learning client archetype analysis
This study was published in the International Journal of Behavioural and Experimental Finance where we found 4 statistically significant behavioural clusters on Momentum Wealth unit trust investors:
1. The Market Timer (switching the most and active in up-risking and de-risking portfolios)
2. The Assertive Investor (active primarily in chasing past performance)
3. The Anxious Investor (loss averse investor who doesn’t mind risk but de-risks at the first sign of turbulence); and
4. The Avoider (risk averse investor who invests conservatively and gets stuck there).
Innovation #3: Supervised machine learning random forest algorithm switch predictor
This study has been provisionally accepted in the International Journal of Behavioural and Experimental Finance and predicts investor switch behaviour based on 40 different features which relate to the long-term risk attitudes as well as short term market movements. Importantly we show the exact values here where each of these features are important. I.e. once someone’s portfolio > R300k, they have switched before, and both their relative and absolute past performance is worse than peers, they are far more likely to switch. In fact each client now carries a predictive switch value which our study has shown to be robust and commercially acceptable.
Furthermore, the predictive model gives a far higher prediction rate for the market timer archetype when previous switch history is hidden from the mode. This shows its efficacy in giving relevant and timely behavioural predictions.
The hyper-personalisation relevance here is that with the personality, money attitude, archetype and behaviour prediction model we have a solid foundation to solve the behaviour tax problem which (since COVID) has cost clients 3.5% per annum on average.
Behavioural track #2 : Invest differently
This is where the product relevance comes in. People’s cognitive and emotional biases make them more suited to certain investment products. Here we published a paper in the flagship annual Behavioural Economics Guide (2023 and 2024) that showed the disposition effect of our local execution-only stock traders on the Momentum Securities platform.
We did this by proving a combination of loss and regret aversion to traders (combined, this is the disposition effect) of 1.4x over the pandemic period. This means that clients with a ratio > 1 trade too much in the gain zone (they are regret averse) and allow their losses to run too long in the loss zone. Remember the investment context is different here.
Innovation #4: Clustering machine learning to show high-risk disposition effect traders
My analysis showed that male Gen Xer’s have a staggering disposition effect of 3.92. Importantly, these traders also incur the greatest behaviour tax – they generate an average return on their trades of 1.5% lower than the JSET rate (rate for keeping cash in their accounts). The call to action here is to implement a stop loss investment strategy. Traders with stop losses have been shown to lower their disposition effect.
Innovation #4: Hyper-personalisation engine
Currently in development is the engine that uses all this personality science and machine learning to give hyper-personal nudges to advisers and clients. This is currently in production and uses an LLM (large language model) with these above inputs to generate hyper-personal messages. For example anxious clients in the anxious investor archetype get a message (when their switch prediction hurdles 80%) that focuses on staying investing and provides support judgement from others that are not switching. The LLM basically hyper-personalises the engagement based on all of these inputs and represents a first-of-its kind hyper-personalisation strategy for any business based on sound personality and behavioural science and powered by AI and Machine learning.
Describe the impact your evolution has had in response to its identified challenges and targeted outcomes.
The main challenge here is in institutionalising and scaling the behavioural finance function for a large investment manager. The intellectual property developed has been recognised by 2 publications in a major international scientific journal (not an easy task for SA researchers) because of the use of massive datasets and publishing robust results. This has also been backed up by publication in the major international behavioural economics scientific trade publication (The Behavioural Economics Guide in 2023 and 2024).
Paul was also invited to present versions of these papers at various local and global scientific conferences:
2022 Economics of Financial Technology Conference (Edinburgh)
2022 World Academy of Science and Technology (Berlin)
2023 International Conference on Economics and Finance (Barcelona)
2023 Association of Computer Machinery AI in Finance Conference (New York)
2024 Academy of Behavioural Finance and Economics (Los Angeles)
The IP developed has been trialled and tested and the next phase is gathering further data on application as this is used in the advice process. The real achievement here is in delivering the intellectual property to power the future of behavioural finance and that is the intersection of psychology (personality science) and technology (AI and machine learning). We believe we are very much ahead of this curve globally given the interest in our work from all spheres.
Supporting Document 1 >
Supporting Document 2 >
Supporting Document 3 >
