Equities for the long run?

Traditional models suggest there is a very high chance equities outperform over the long term. But are they overconfident, and should long-term investors adjust their asset allocation accordingly?

• Many investors, including defined benefit (DB) and defined contribution (DC) pension schemes have time horizons extending far into the future. As such it is crucial to understand risk and return trade-offs over the long term
• Traditional models can be poor indicators of long-term risk because they do not allow for the risk that the assumptions used, particularly expected returns, could be wrong
• Allowing for this uncertainty promotes maintaining a healthy level of diversification across asset classes, even over multi-decade time horizons. It also suggests that DB and DC schemes may require higher levels of contributions to remain confident of achieving their long-term objectives.

Understanding uncertainty is critical to making good investment decisions. Stochastic models (‘stochastic’ simply meaning they use probabilities) are widely used in risk management to help assess the likelihood of different outcomes over different time horizons.

Whilst such models should never be used blindly, and one needs to be careful about ‘driving only using the rearview mirror’, they can help investors understand the risk and return trade-offs they face both in the short and long term. Importantly, they lend a degree of objectivity and help avoid a natural tendency to bias towards a particular viewpoint at the expense of historical precedent – a behavioural effect known as base-rate neglect1. Ignoring base-rates – and starting from the premise that ‘this time is different’ – is dangerous and gets many investors into trouble.

However, a problem with traditional models is that they can paint an overly confident picture of what the long-term looks like and the asset classes that will do best. This can lead to over-aggressive or under-diversified strategies, and contributions levels that are too low. Happily, there are relatively straightforward steps that can be taken to improve these models: in this paper we explore the implications of doing so and arrive at some interesting conclusions.