The low-volatility anomaly basically says portfolios of low-volatility stocks have produced higher risk-adjusted returns than portfolios with high-volatility stocks in most markets studied. Now often most of these low-volatility anomaly studies take one of the following two approaches -
Ranking-Based Approach:
In rankings based approach, the market or target segment (like large cap, small cap, emerging etc) is divided into deciles/quintiles based on a volatility measure. The division is such that securities in the lowest decile/quintile will be of low volatility. The portfolio is then invested in these low volatility deciles (or weighted heavier) and re-balanced monthly.
Minimum Variance Approach:
Another scheme is constructing minimum variance portfolios with the understanding minimum variance portfolios will have lowest risk. Then a weighting algorithm is used to determine the weights and limits for the selected securities & sectors belonging to that minimum variance portfolio. Then the portfolio is re-balanced monthly.
While the idea is good, I am not sure either of the above approaches are practical for individual traders unless one has large account and time. Also my personal preference when it comes to academic papers on trading is to generally pick the concept, understand the authors viewpoints, discard rest and figure my own way to incorporate those concepts for profitable outcome.
Capturing Volatility Premium:
IMO often good ideas come from simple rearrangement of concepts picked in various contexts over time. Applying that here, what do we know when it comes to volatility and these approaches -
- Volatility in markets is mean reverting i.e., low volatility begets high volatility and vice-versa.
- A big part of low volatility portfolio returns is due to periodic re-balancing of the portfolio.
If that is true, then why not simply pick few broad market (liquid) ETFs, buy when their volatility is low and sell when their volatility is high?
Let's put above hypothesis to test. The broad market ETFs chosen for the test are - Emerging Markets (EEM, Europe (EFA), Asia & Pacific (EPP), US Small cap (IJH) and US Mid cap (MDY).
Some Notes:
- Average True Range is used to measure volatility here. There are other ways to measure volatility. The choice of ATR as volatility measure is mostly a matter of convenience.
- The test results are frictionless i.e., no slippage and commissions.
- The test is done on weekly charts. Duration: 2000 - Current.
- The portfolio is weighted equally across the 5 major markets.
Following annotated images provides various performance stats. One can glean several insights both at individual market level as well as at portfolio level. Some highlights:
In the below image, notice the horizontal areas in the equity curve (black line) and the behavior of benchmark (red line) in those periods. The system goes into sidelines or has position only for a short time when the volatility is high in the benchmark. That is what we want.
The following image provides various performance stats and ratios both for individual markets and for portfolio. The pie-chart provides the color notation. Notice anything of interest in "Annualized Sharpe", "Sortino Ratio" and "Rolling Correlation" bar plots?
The below scatter plot shows where individual markets and equal weighted portfolio fall in annualized Risk-Return spectrum.
The last image provides detailed performance stats, calendar returns and draw downs etc. Looks like US Small cap has better returns of all whereas on risk-adjusted basis, the account seems to do better.
Now one idea doesn't make a system. The purpose of the test is basically to check for myself whether the hypothesis (i.e., buying and selling based on volatility level and price action )has legs and worth investigating further. The results are better than I expected for first round. The hypothesis seems to be worth investigating further. Thoughts?
I have not seen any low volatility anomaly studies on net that approach it this way. If you know of any studies/articles that discuss low volatility anomaly using approaches (besides ranking into deciles or using minimum variance) then please let me know.
Side Note: Like other tests on the blog, formulation of test rules, back-testing, analysis and visualizations are done using a proprietary software I developed over time. The software was built using R language and C#.
Wish you all good health and happy holidays!
3 comments:
Hi, how do you decide when to sell? When ATR> than some fixed treshold?
Hi Jozef,
Thanks for reading the post. Basically I used rolling ATR and converted it to act like an oscillator between 0-100. 0 is low and 100 is highest. Sell is when value > 80.
Regards
Hi,
This is indeed quite intriguing. Out of curiosity, how did you manage to convert ATR to an oscillator between 0 and 100?
Post a Comment