Home > How Exponentially Weighted Moving Averages Work

How Exponentially Weighted Moving Averages Work

We use exponentially weighted moving averages (EWMAs) quite a bit in a few algorithms. They’re one of the tricks everyone should have in their toolbox for cheaply approximating recent history of a time-series variable. We even wrote an open-source implementation in Go. How do they work? That can be a little confusing, but I’ve found that pictures are a good way to explain it. Here’s the pictures I use. First, draw your time-series as a bar chart. Exponentially_Weighted_Moving Now initialize your EWMA. There are several ways to do this, but one way is to just set it to the value of the first bar. We’ll color-code this one dark red. Exponentially_Weighted_Moving_2 Now each further point in the time-series gets a combination of the current value of the EWMA, plus the new value, which we’ll color-code with a unique color each time. For simplicity, we’ll set the decay factor to 50%. This means each time we add a point, we’ll decay the existing EWMA by 50%, and add 50% of the new one to it. You can see that the new EWMA consists of half the first red EWMA, and half the second black value from the time-series. That means it’s closer to the second black value than it was previously, but it’s not all the way there: Exponentially_Weighted_Moving_3 Now just rinse and repeat. Notice the way that the EWMA tracks the general trend of the time-series value, but smooths out its extremes. Notice also that each value’s influence diminishes exponentially over time. Exponentially_Weighted_Moving_4 Are EWMAs the best thing since sliced bread? Well, they certainly have their uses, but obviously they don’t fit all possible cases.
Baron Schwartz
Baron is a performance and scalability expert who participates in various database, open-source, and distributed systems communities. He has helped build and scale many large,…
Read more

Tweets

SolarWinds's Twitter avatar
SolarWinds
@solarwinds

Digital experience monitoring, infrastructure and application monitoring, and log management—all from a single pane… t.co/StlW5tJ6Ey

SolarWinds's Twitter avatar
SolarWinds
@solarwinds

SQL server performance problems? Try these five troubleshooting tips to get back on track. t.co/nPvYVjHTT9

SolarWinds's Twitter avatar
SolarWinds
@solarwinds

Listen to what we’re doing in response to the SUNBURST cyberattack, and what you can do to raise your IT defenses o… t.co/M9JXJIRFEn