Volatility is one of the main ways we describe risk in the managed futures world, and it’s reflected in the calculation of several other measures of a CTA risk/return profile (like Sharpe Ratio). But last November, Newedge released a paper arguing that the typical method of calculating volatility is flawed because it leaves out an important factor: autocorrelation. Recently, Futures and Options World summarized Newedge’s findings in a more layman-friendly write-up, so we thought it would be worth revisiting the topic for those who wanted the plain English version. (Or you can read the original Newedge research here).
Typically, volatility is calculated by multiplying the standard deviation of a CTA’s returns by the square root of the time series. In other words, if you’re looking at a monthly time series, you would multiply the standard deviation by the square root of 12 (the number of months in a year) whereas if you had a weekly time series you would multiply it by 52 (weeks in a year) and for a daily time series you would use 252 (roughly the number of trading days in a year). This gives you the volatility in percentage terms.
Newedge’s research argues that the drawdowns CTAs experience don’t always match what we would expect based on their volatility, and they point to autocorrelation as the missing piece of data. You see, the typical volatility formula assumes that one period’s returns are independent of any others – that whether a CTA made or lost money in one month will have no bearing on whether or not it makes or loses money in the next month. But when a CTA exhibits autocorrelation that assumption no longer holds true. Positive autocorrelation means that whatever happened last is more likely to happen again (winners win and losers lose) while negative autocorrelation means that whatever happened last is less likely to happen again (reversion to the mean).
And according to Newedge’s work, trend following CTAs tend to exhibit negative autocorrelation. (Just more reason to listen to our advice about the benefits of allocation during a drawdown, although of course past performance is not necessarily indicative of future results). So for most trend following CTAs, Newedge’s work would suggest that we are consistently overestimating their effective volatility (and underestimating it for managers who exhibit positive autocorrelation).
We won’t be switching over all of our volatility calculations to incorporate autocorrelation just yet, but it’s definitely something to keep an eye on.