The big innovation we have come with, at Bramham Gardens, is that Risk is not well characterised by volatility, especially at a time when it is heavily controlled and kept under a lid.
In fact the idea that one measure of univariate risk is sufficient to characterise the Risk overall in the market is deeply related to the idea that the market movement can be synthetized as related to one single Representative Investor. We think it is not the case. The community of investors is deeply heterogeneous more often than not and when it becomes homogeneous, this shows and it is worth noticing.
In Riskcasting®, the way we look at risk is to understand how different market segments (i.e. investors taking different optional bets) look at risk differently. We have found the relative evolution between these various segments to be very informative. In addition, The overall Finding is that we have been able to observe lasting risk regimes, although we reassess our Understanding of Risk on a daily basis. As a result, the number of risk shifts per year is limited to 9 / 10 per year. This means that coincidence is not an issue here.
In Low-Turbulence®, we consider fixed time windows and look at segmenting the price signal among various frequency groups. In each frequency group, we observe a degree of energy, or by analogy a level of volatility. This vector of energies / volatilities per frequency group gives us a signature of the environment we are in which is very rich. We then reduce dimensionality in order to really get the gist of the signal using techniques well established to migrate pictures from high to low resolution, while keeping an optimal level of similarity.
These approaches differ from PCA in the sense that PCA still does not see Risk as more than univariate volatility.
Our core finding is that Risk is multidimensional and not well characterised by the traditional moments (volatility, skew, kurtosis,…), because nobody knows how to properly ponder these moments.
Here is the economic intuition: by looking at Risk in a comprehensive manner we get a much better grasp at it, than when looking at a small subset of risk, i.e. volatility. The second point is that the overall risk attitude in the market, its “mood” is not white noise or stochastically random at a relatively high frequency. There are lasting trends worth capturing beyond the daily mean-reversion of prices.
It should be noted that the daily reassessment of the market mood is not supposed to automatically trigger action as a result, but to have a proper indicator of when risk attitudes start to shift, without having to wait for any artificial fixed steps such as end of week / end of month, etc.
In Riskcasting®, we really have some predictive power embedded, with a hit ratio in the region of 55%, and a high upside to downside capture ratio, which is unusual for such a broadly traded index like the S&P500.
Low Turbulence® has slightly less predictive power, but does a very good job at removing the systematic component of risk. In addition it is implementable on all series of prices, whether indices or single stocks. This means that when blending different combined equity/bond indices per region, we obtain noticeable strong results, having diversified the idiosyncratic components.
Last point, although the measures of risk are more refined than what most people typically use, the implementation is extremely simple, as it entails rebalancing two assets per strategy a few times a year.