"Is three year volatility a good risk measure?" is not the same question as "is three year volatility a good predictor of volatility?’" Or as many people might frame the question: "is today’s measure of three year (historical) volatility a good gauge of future risk?"
Definitions of risk are multivariate, personalised and subject to changing circumstances and outlook. However, volatility, as measured by historic standard deviation, is one of the most commonly used and standard measures of risk in investment. Yet, even such a simple measure has to be understood to be useful.
For multi-asset investors there are extra complications in measuring and estimating risk, which are two very different processes. One has to contend with many individual investments; not only understanding their changing risk characteristics, but also the ongoing dynamic of how these investments interact with each other – how they relate to each other and how they correlate.
We will keep this discussion mostly focused on one asset class. It is important to understand the nature of volatility to help us decide whether we should focus on any one measure and how it should be ‘read’. Adding other investments, non-correlating or otherwise, complicates things further and may be looked at in more depth in another article.
The following should be seen as a general discussion, as volatility characteristics differ from asset class-to-asset class and change through time as their composition and valuation changes. Illustrations and statistics used are from weekly FTSE All-Share Total Return data from January 1999 to date. The broad behaviour of volatility through time is relatively consistent even if the measure itself is not stable. This is the first and most important concept to understand.
Volatility is not stable
Although we understand short-term volatility is in itself volatile, most do not understand that even three year rolling (historic) volatility is also quite unstable. As such we may wish to conclude that three year volatility is unlikely to be a good measure of risk – although this does not mean there is no information surrounding this conclusion.
Rolling three year volatility of FTSE All-Share Index
Historical risk, as measured by annualised standard deviation of returns, is relatively easy to calculate. We also understand that there can be a big difference between the risk that was taken by investing in an asset class and the ‘risk’ that was experienced during any particular period. However, can we use historical realised volatility to tell us anything about the future risk we are taking?
The following graph looks at the relationship between historical three year volatility and the (average) forward realised volatility relative to that previously experienced. The x axis is historical three year volatility; the y axis is the change in volatility over the next three years.
Relationship between historical three year volatility and the forward realised volatility
Despite volatility being used synonymously with ‘unpredictability’, it does have some consistent characteristics, of which, mean reversion is one. As we can see from the above chart, periods of higher risk tend to follow periods of low risk and vice versa. For example periods that have experienced say 9–11% volatility, on average, experience volatility of 20 to 25% in the next three years (versus a period average of 17.5%).
Understanding this may help investors interpret today’s measure of risk. Yet if you knew the longer term average risk measure and ‘today’s’ measure of risk, should you adjust your assumption for the risk you are likely to face? The chart above suggests that if you are currently sufficiently away from the longer term average – then you should.
To help us frame our expectations we would therefore require today’s measure,
the longer term average and perhaps the historical high and low of this measure. As well as the understanding that not only is volatility mean reverting, but that lower than average realised risk tends to be followed by higher than average risk and vice versa.
The answer to the original question we asked – ‘Is three year volatility a good risk measure?’ – is therefore nuanced. There appears to be information available to us as investors as long as we do not take the latest reading as the sole estimate for future risk.
For many investors risk is defined and coded as the potential to lose money rather than purely experience volatility.
Having looked at historical volatility as a predictor of future volatility, are we able to look at historical volatility as a predictor for future returns? Again, I should stress the limited application of this finite data interpretation; although the findings are intuitive given the nature of what we are using as the underlying data i.e. UK Equity return data from January 1999 onwards.
The following table shows future three year annualised returns after segmenting the data into different levels of historical volatility:
||Realised three year annualised returns
The average during this period, for three year annualised returns, was 8.7%.
Given that high volatility tends to be associated with falling markets and low volatility with rising markets – it is not surprising that after periods of high volatility, future returns will likely be higher and vice versa. This does not suggest one should time markets, but that expectations should be adjusted by recent experiences and not extrapolated.
Combining these elements suggests that risk is dynamic. While there are many different measures (and different focuses for different investors), even the simplest of them require understanding and context in order to be useful and helpful guides for future risk.
For multi-asset portfolios, the picture gets more complicated as one introduces the concept of differing return characteristics (across the different asset classes) and varying correlations between them. This requires an extra level of understanding and potentially more risk measures. One noticeable improvement has been the use of a standardised measure. For example, giving a risk measurement as a percentage of equity risk, assuming all aspects mentioned above are recognised and understood.
If we added UK gilts to UK equities we can start to illustrate some of these issues.
Given the varying correlations and our understanding of how we expect UK equities and gilts to behave, we might expect a far less risky outcome.
However, the following chart shows the relative risk of an 80/20 equity/gilt portfolio (portfolio risk divided by 100% FTSE All-Share Index Total Return risk).
Relative risk of an 80/20 equity/gilt portfolio
On the one hand we might be disappointed that the diversification benefits are not as great as we might have estimated them to be (average is 78% with a high/low of 75.9%/80.3%). However, used as a risk measure it is reasonably stable – allowing us to understand the strategy risk by proxy to something else.
As the relative weighting of ‘other’ assets increases, this measure understandably gets less useful as this stability decreases. For instance, if we look at a 50/50 portfolio, the average/high/low (over the same January 1999 to present period) is 47.4%, 42.0% and 53.2% – i.e. the variation in the relative measure has increased from 4.4 % to 11.2% (high – low range).
The driver behind this measure’s stability is the degree to which the volatility of the numerator (the portfolio or strategy) changes in relation to the volatility of the denominator (FTSE All-Share Index Total Return risk). The more stable or more diversified the strategy when compared with equities, the safer it will look during ‘riskier’ times and the riskier it will look during ‘safer’ times. This is because you are dividing by an equity measure, which typically varies the most and is currently towards its lows.
As in the earlier discussion on single equity risk measures, the more unstable the measure, the more we should look beyond the current measurement to historic highs, lows and averages. We should also understand that these measures are more likely to revert towards their historic averages. Just as when pure equities appeared to be safe (on historical risk measures), they turned out to be more risky going forward. A riskier multi-asset strategy (as measured on current basis versus equity or another equity heavy strategy) may also be safer looking ahead.
Similarly, we may ask the question: if two assets, strategies or portfolios have the same three year risk measure, are they equally risky? Hopefully, by now we know that we need more information before we can answer. If, for instance, a diversified multi-asset portfolio has the same three year standard deviation as an equity dominated portfolio currently, due to recent low levels of equity volatility, we might suspect that the latter portfolio is likely to be more risky. Extra measures such as historic average risk scores, or other measures such as drawdown, would help further our understanding as to relative riskiness between different assets and strategies.
Risk requires a good understanding of underlying asset classes and investments and is usually better interpreted with several measures – here we have barely covered a couple of simple ones – especially if one is trying to compare across different investments or strategies.
Not all measured risk is the same. Generally it is not stable and we would be better off thinking of volatility as being mean reverting rather than extrapolating. As such three year volatility may provide information, but should not be used in isolation as the sole measurement of portfolio risk.