#sigmaspikes

We have limited time and attention. This is one of the realities that developing traders soon face—no matter how hard you work or how well-intentioned you are, there’s a limit to what you can look at, analyze, and understand. It’s important to filter out the noise so we know where to focus that precious resource of time and attention, but how do we do that?

Another way to ask this question is something like “what’s the most important thing on a chart?” or “where do you start reading a chart?” The answer is simple: you want to pay attention to the big moves.

Why big moves? Because most of what we see on a chart is random. Most of the time, the market moves unpredictably from point to point, and we can’t make money if the market is moving randomly. If we’re going to risk our money (and other resources like time and mental and emotional energy), we need to have some evidence that there’s an edge in the market. Looking around big moves is a good way to start looking for an edge.

So what is a big move? The answer is “it depends”, but that’s not the useless answer it might appear to be! What does it depend on? It depends on how the market usually moves; a more precise way to say the same thing is that it depends on the volatility of the market, and that’s the key: using measures that automatically adjust for the volatility of the market can help us find these big moves we want to focus on.

We can think of “big moves” either as single bars or swings—we can look at large single bars, relative to some measure of “average bar size” on the timeframe we are monitoring, or we can also look at large swings that somehow rise above the average swing length for the

The first is visual inspection. Simply look at the chart and ask yourself which moves jump out as big moves. The problem with this is that this is a purely subjective measure. What stands out to you today might not stand out tomorrow; if you look at the same chart twice you might come up with two different answers. That answer might depend on the charts you look at before, how long you’ve been studying, what you had for breakfast—you get the idea. This certainly doesn’t invalidate this approach, but you do need to realize that it may not be fully reproducible. The chart below highlights some big moves with red arrows, but the orange arrows point to bars that might or might not be considered big moves. We also can’t use this in any kind of backtest because it’s so subjective. Even with those limitations, this is something anyone can start doing today and simplicity is often a good thing.

We can pick out big bars just by looking at the chart, but this raises some questions

If we want to have something more consistent, we can start applying some mathematical measures to charts. One of the tools I use is something I’ve called SigmaSpikes, which is simply each day’s move expressed as a standard deviation of the past 20 trading days. One way to use it is to go through a chart and highlight all the moves that exceed a certain threshold value. (I often use 2.0) The beauty of this measure, to my eye, is that it lines up very well with the points I would pick out subjectively, but it’s an objective measure. I can scan charts for it, create screens that rely on it, look at markets that have made big moves in recent history, and use it in a backtest—all good tools to help me narrow down the universe and to know where to focus my attention.

Big bars measured with sigma spikes

Another useful tool is to put some kind of band on the chart, and to look at the spots where the market engages the bands. Personally, I prefer Keltner Channels, but Bollinger Bands are probably more common. Don’t get too caught up in the search for the perfect tool or perfect parameters; the concept is much more important. The concept here is that we put some volatility-adjusted bands around the market and only look at the point where swings (or single bars) rise to the level of significance by touching the bands. Again, we can look at historical measures of this objective tool and even incorporate it into backtests and systematic approaches.

Bands can also be useful to measure big swings

Many of my readers probably already use tools like this, but, if you don’t, think about adding this concept to your toolkit: focus on markets that have made big, sharp moves, as these moves may point to the way to some imbalances in buying and selling pressure that could give opportunities for profits.

I’ve received a few questions about how to calculate a measure I have referred to many times here on my blog and in my published research that I call SigmaSpikes. I’ve disclosed the calculation in several places, but the question keeps popping up, so maybe I need to do it more clearly! First, a little history on how I came up with the measure, or, well, reinvented the wheel:

I think most of us who are involved with markets for any length of time begin to realize that volatility is important. (Yes, that statement is a dramatic understatement.) I did a lot of my early semi-quant work in a vacuum, so I didn’t know what everyone else had already done; I created really cool and useful tools that were–ultimately–things that everyone else was already using. Though I kind of smile at my naivety now, I think that process was an important part of my development and learning, so it wasn’t wasted time.

When I began to think about volatility, I came up with the idea that different markets traded with different volatility levels (true), that it would be a good idea to have some way to compare volatility across markets (very true), and that each market probably had a base volatility level I could expect it to return to. (The last idea is not true in any practical or meaningful way, but two out of three wasn’t bad.) My first attempt was to look at each day’s move as a multiple of the average range of the market over some time window. I played with different time windows and decided that a month was about right for how I traded–any longer, and it would be too slow to respond to any change in volatility, and much shorter was so unstable that it wasn’t useful. I settled on 20 trading days as a nice round number and used it for my baseline calculations.

The next evolution was to use true range instead of range. This is a small difference, but it really is better: for one, it automatically catches overnight gaps and considers them in the volatility of the market. Another point that is often overlooked is that ATR allows us to use “close only” data series in range calculations, and I was also beginning to use some economic data in my work that had no range associated with each period, just a single datapoint.

Here’s an important point, and it’s probably the most complex of this post: “time leakage” is a real problem in any kind of testing or development. If you create a system and even slightly reference a seemingly insignificant datapoint you could not have had at the time, your system results will likely look very good, but they’ll also be completely invalid. We must be very careful to only use data we would have had at the time. The best way to measure this volatility ratio seemed to me to measure against yesterday’s ATR, because we don’t know what today’s was until the close of the bar. A very big day today will inflate the baseline, making that big day seem smaller, by comparison, than it would have been to a trader at the time. So, I offset the calculation to measure today’s range against yesterday’s 20 day average range.

This question frequently comes up on trading forums: a “return”, in finance, is just a percent change. This is standard language in any academic or mathematical finance, but not so much in the literature of technical analysis. A return is simply a percent change, so a daily return is the % change for the day.)

At some point, I began to think that returns were a better way to think about market price. Once you make that leap, standard deviation of returns is an accepted measure of volatility, and is trivially easy to calculate. So, I settled on the idea of looking at each day’s return as a multiple of the 20 day standard deviation, with the measure offset one day, and that’s what I use today.

Since then, I discovered that practically every options trader and risk management system in the world uses this exact calculation (or some slight variation) in their work. It’s a very useful measure, but is by no means unique to my work or perspective. It is, however, a very useful took for looking at market action and price changes, quickly assessing volatility, and removing much of the emotional component associated with large price moves over a period of time. More on that, perhaps, in a future post.

There are a number of ways to do this calculation, but here, in bullet points, is what you’re trying to do:

Calculate daily returns as: return = today's close / yesterday's close - 1

Calculate standard deviation of daily returns over a 20 day window. In quasi-Excel langauge:: stdev = stdev(returns, 20)

Calculate today's sigma spike: SigmaSpike = today's return / yesterday's stdev