How not to measure risk tolerance

Over 8 years ago in 2008, in the early days of setting up the Barclays behavioural finance team, and just before we were hit by the financial crisis, I wrote an article on things to avoid when designing a measure of Risk Tolerance.

It contained 12 principles that designers of such tools should follow to ensure that what they are measuring provides a valid input into a suitability process. The Risk Tolerance measure we designed using these principles differentiated very effectively between individual investors and, crucially, proved to be extremely stable even through the market mayhem that followed.

This article should now appear dated, with the principles below as nothing more exciting than the industry standard. Gratifyingly there are now a number of well-designed Risk Tolerance measures available. However, there are also numerous measures still in use that fail to meet many of the basic standards laid out in the Principles below – so they bear repeating in 2016.

Most disturbing is a trend that purports to be an advance in Risk Profiling, but is actually a dangerous step backwards. This is the excitable desire to use the shiny new toys of big data, and artificial intelligence to determine a client’s ‘risk profile’ from their observed behaviour – ‘revealed preferences’ in the technical parlance. This fails on principles 2, 6, and 8 below.

The whole purpose of measuring Risk Tolerance is to establish investors’ stable long-term preferences for trading off risk versus return. And yet, most actual behaviour is short-term, extremely unstable, and highly influenced by framing, context, and biases. It is vital to understand these short-term emotional responses to markets… but as a route to helping them overcome myopic actions, not to bake these into a recommended investment solution. Revealed preferences should absolutely not be used as a foundation for Risk Tolerance – you will end up ‘optimising’ a solution for investors based on aspects of their short-term behaviour that they should strive to control, not pursue.

There is an important role for using sophisticated technology and data analytics to extract patterns of behaviour, and help to guide investors to better behavioural. But using these tools to assess Risk Tolerance over-engineers a misguided solution for a problem that doesn’t exist.

For a more thorough and sophisticated discussion of how Risk Tolerance should be thought of, and in particular why it is also misguided to try to apply it to sub-components of a client’s wealth (as is often the case in ‘pot’ based approaches to goals-based investing), see Risk Tolerance: Essential, Behavioural, and Misunderstood.

Principles – what not to do:

  1. Do not confound risk tolerance with investment objectives
  2. Do not confound with other aspects of risk attitude, behaviour or other personality dimensions
  3. Do not require respondents to have knowledge of finance or investing
  4. Do not require respondents to perform numerical computation or probabilistic reasoning
  5. Do not require respondents to have knowledge about current market conditions
  6. Do not confuse past behaviours with optimal actions
  7. Do not rely on future beliefs or expectations more than risk tolerance
  8. Ensure that the measure is stable over the market cycle
  9. Only use questions that discriminate effectively between individuals
  10. Avoid ‘social context’ questions (questions to which respondents may feel uncomfortable responding honestly)
  11. Avoid multi-clause statements
  12. Don’t guess what questions to include – use rigorous statistical and psychometric techniques