Blog Posts — March 1, 2018

The Increase in Volatility? Yup, It Was Big

by Melissa R. Brown, CFA

While the US market has retraced some of the path from its recent downturn, it’s a different story with market risk, which bottomed out in late November last year, and rose sharply since, especially in February (Figure 1). What’s more, we believe that looking at close-to-close performance (as we do in our risk models) has masked some of the heightened intraday volatility, which likely has had a further impact on portfolio performance.

The data behind this assessment reveals that we have been through an unusual period. We looked at three measures: 1) the spread between the daily high and low values of the Russell 1000 divided by the opening price; 2) the high-low spread adjusted for the expected level of daily volatility that day, using Axioma’s short-horizon US4 fundamental model; and 3) the five-day change in the short-horizon risk forecast, and we compared the daily spreads to daily returns to see if spread volatility had, in fact, increased.

For the month of February, the daily high-low spread averaged more than 50% higher than the average since 1992 (2.0% vs 1.3%). The difference is even bigger when adjusted for the expected risk (2.9 vs. 1.3), the result of relatively low risk expectations at the beginning of the month. Histogram distributions of the percent of days the daily spread fell into a specific bucket show a wide skew to the right, with two of the worst[1] days in history occurring in February 2018 (the 5th and the 9th, Table 1).

Next, we looked at whether there was a true jump in volatility of the high-low spread vis-à-vis closing prices. We calculated the ratio of the spread to the daily return for the 1861 days since 1993 when there was a big daily return, defined as a more-than-one-standard deviation event based on our short-horizon risk model. By definition, spreads are almost always higher in magnitude than daily returns[2]– the ratio of the two averaged of 1.25 higher on big down days and 1.14 on up days. If there was no true difference in the underlying volatility of the two (so spreads were not statistically higher than daily returns), the ratio would be expected to exceed 2 less than once a year (and be greater than 2.65 once a century!) The distribution over the full period shows that on the big down days (the left set of bars), February was no different from what would be expected, but when the market was up strongly, the high-low spread did seem to exhibit higher volatility than the daily price change. We reached the once-a-year ratio (2.14) on February 6, and the once a century (2.79) on February 9. Clearly, therefore, volatility of intraday spreads has increased more than that derived from closing prices, and this is in addition to the sharp increase in overall volatility.

Since February 1, the short-horizon fundamental risk forecast for the Russell 1000 has risen fairly steadily from just under 8% to more than 13%. Risk changes (in percentage-point terms) over rolling five-day periods since February 1 have also been unusually large when compared with history. Five of the 60 biggest five-day risk increases since at least 1993[3] occurred in February, with the biggest one-day moves on February 5 and 8. Figure 3 shows the historical distribution of five-day risk changes, and February 2018 clearly stands out with many more increases, and far more very big increases.

We started this short study thinking that perhaps the persistently low levels of volatility we observed in 2017 made it seem that the recent jump in volatility was higher than it actually was, but that was not the case. Volatility measured intraday or close-to-close has taken an unusually high leap. While it may settle at these levels for a while, we see no evidence it is likely to go back to levels of late last year any time soon.

[1] We assume that big intraday swings are not good for investors 
[2] They can theoretically be lower if the market opens up from the prior close and never dips below the opening price but that is relatively rare
[3] Out of 6345 five-day periods