Low Volatility vs. High Volatility Days


I read a blog post recently that began “suppose you have a trading system that works well on low-volatility days…” and I thought, hmm. Is that a thing? Is there an edge to low-volatility days vs high volatility days? Let’s turn this blog post into a speculator’s version of Dude, What Would Happen?

The parameters:

SPY, baby! From 2000 through 02/16/16.

• A low volatility day is defined as a less-than-1% change (up or down) from the previous close to the current close. A high volatility day is greater than 1%. Presumably, days that move exactly 1% to the umpteenth digit are falling through the cracks. Call me lazy.

• Gain/loss is recorded from the current close to the close of the following day. This would be a long-only system. This is unicorn-and-rainbowland so there are no commissions or fees of any sort.

That’s it! Not a trading system. Just a filtering system. Let’s take a look:

Screen Shot 2016-02-17 at 2.09.39 PM

Screen Shot 2016-02-17 at 2.13.47 PM

Those are the graphs for all low-volatility days as defined above. There’s a positive expectation but only just. Most of the time any system built around this would have been underwater. In fact, this looks suspiciously like a buy-and-hold curve from the same period. No thanks!

Screen Shot 2016-02-17 at 2.14.12 PM

Screen Shot 2016-02-17 at 2.14.21 PMWell now that’s interesting. The cumulative equity graph for high-volatility days spends most of its time above zero. That’s ALWAYS a plus in my book (hah, I just made a math pun there…). This curve looks great, except for a couple of times where it looks really awful. But still, better than the low-volatility days.

Right, there we have it. Only trade on days when there’s been a lot of movement in the market, and hope for the best.


What if we divided up these low- or high-volatility days into up or down days? Up is when the close is above yesterday’s close (rocket science!) and down is…ok please don’t make me actually spell it out for you.

Screen Shot 2016-02-17 at 3.04.29 PMScreen Shot 2016-02-17 at 3.04.39 PM

Low-volatility up-day trading looks like garbage.

Screen Shot 2016-02-17 at 3.06.32 PMScreen Shot 2016-02-17 at 3.06.46 PM

Low-volatility down day trading is less sucky but still sucky. Ok now on to high volatility days:

Screen Shot 2016-02-17 at 3.05.51 PMScreen Shot 2016-02-17 at 3.06.01 PM

Wow! Yikes! It would almost be a good short-selling idea, except most of the gains (or losses, in the above graph) happened around 2008-2009.We certainly can’t spend our days wishing for the Crash of 2008 again.

Screen Shot 2016-02-17 at 3.05.11 PMScreen Shot 2016-02-17 at 3.05.23 PM

Ooh! Now we’re getting somewhere! (Isn’t it weird how the good results always seem to appear at the end of my blog posts?) Trading on high-volatility down days shows a much more consistent equity curve. Drawdowns are less severe, and the differences between bull and bear markets are much less pronounced compared to the other systems. This system does tend to stall out during bull markets compared to more turbulent times, but it’s still kind of impressive.

These guaranteed can’t lose trading systems exploratory ideas don’t all trade with the same frequency. The high-volatility-down-days system trades much less than the all-low-volatility-days system. We should be careful comparing them directly. However, a quick and easy way to compare them is to calculate the average gain/loss per day. That equalizes systems with different trading frequencies. Here’s a graph:

Screen Shot 2016-02-17 at 3.11.18 PMSo yeah. High-volatility down days are the way to go.

Which I guess just proves that the market is highly mean-reverting in the short term, and highly up-trending in the very long term.

12 thoughts on “Low Volatility vs. High Volatility Days”

    1. Yeah good idea! Something as simple as looking at some multiple of average true range, and using that as a threshold to filter. Lots of different directions one could go with this. Thanks for your comment!

  1. Hi Matt,

    Thanks for this post – very insightful indeed! It kinda supports that I read once: short-term speaking, bear markets are mean-reverting and bull markets are trend-following.

    But there’s something I don’t understand. What are the buy/sell rules? For example, for the “High-volatility down days” strategy, do you buy when SPY closes down by more than -1%, buy at the next open and sell at the next close?


    1. Ho Tonio and thanks for your post. The buy rule is to buy at the close of the actual day being measured. I.e. if we’re looking for >1% down days, we buy at the close when the close has a greater than 1% loss from the previous day. If you’re then thinking about look-ahead bias with that statement, please see my response to gorynyich’s comment. 🙂

        1. Yes, you’ve got it. Buy at the close of the signal day, sell the following day’s close. Not that it’s a system, mind you. That’s just the parameters my research used.

  2. Interesting graphs, but unfortunately your research have forward-looking bias… You just can’t observe close (to decide whether it is up or down day, volatile or not day) and trade on it at the same time. More honest is to assume that you can trade only at the Open of the following day at best.

    1. Hi and thanks for your comment! I’ve heard this argument before, but I disagree. Yes, technically you can’t know the closing price until the close has already happened, at which point you can’t trade on it. You can however calculate ahead of time what the closing price needs to be so that it’s above or below your threshold. The vast majority of days, you know that the close will be well to one side or the other of your threshold. You simply then set market-on-close buy order, or a limit-on-close buy order if your broker offers it. If it looks like it’s going to be close to the 1% threshold (or whatever you’re using), you can just place a market order near the close, and be a few cents off.

      This means that the results will be slightly ‘fuzzy’ compared to the backtest. They won’t be *wrong*, but they won’t be exact. “Close enough” is the technical term. 🙂 Some days, you’ll get a buy price that is a few cents one way or the other from the close. And some days, you might end up buying and the close turned out to be a few cents in the wrong direction from your threshold. However, 0.99% vs 1% is probably not going to make much difference. Especially since we don’t know if .99% really was the magic threshold that day anyway.

      Even if you don’t agree with my methodology, the research can still be a springboard for your own systems development. Good luck!

  3. …I’ve done similar analysis (fyi latest post on my blog) with same conclusion: down days are likely to mean-revert. That is kinda common-sense, so the challenge is to make a tradeable strategy out of it. I’m sure you’re trying to crack the same puzzle, so here is an idea that I’ve had in mind for some time, maybe it could help:
    Actually trading should be a no-brainer if we know what market regime we are in, assuming there are just two : low-vol-uptrend and high-vol-downtrend. The challenge is of course determining in which regime the market is operating, and that is a really tough nut to crack. Wonder if we could use a ‘supervised learning’ classification based on previous data to determine the maret regime. You’ll need to label by hand the historic data periods being either bull or bear. Then create a training set consisting of possible indicators (volatility, vix futures curve, previous returns etc) and try to train a classifier that determines the market regime. This is a classic pattern-recognition problem. I’m sure many people have tried this, but I haven’t come across meaningful results on in the blogsphere. This can of course mean two things: either it’s just too difficult, or the results are too good to share ;-). I’ll certainly dive into the matter myself.

    1. I’m sure I shouldn’t be reading this at 7am on a Sunday morning. I know I won’t be able to contribute anything meaningful. 🙂 Except to note that yes indeed, that sounds worthy of pursuing further. Saving and flagging your comment for further review. Thanks!

  4. Great post Matt!

    Ernie Chan’s post on predicting volatility ties closely in with this and marrying the two ideas is super interesting. The idea being if we are in a bear market (pick your favorite method of determining that) and GARCH (or some other method of forecasting volatility) predicts an uptick in volatility the next trading period, we short with no mercy. See Ernie’s blog post from Nov 2015 on volatility, GARCH and VXX. If anyone enjoyed this they will enjoy Ernie’s post.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.