What time constants does an EWMA training-load ratio use?

A 7-day time constant (lambda about 0.25) gives a fast-responding acute read; a 28-day constant (lambda about 0.07) gives a slow chronic read. Dividing the acute EWMA by the chronic EWMA gives the workload ratio (Williams et al., 2017).

Does an EWMA workload ratio predict injury?

No. Smoothing the math does not rescue the injury-prediction claims that made the workload ratio famous; that research has been seriously challenged. Treat it as a cleaner description of your recent loading, useful for pacing a return, catching an accidental ramp, and confirming a deload.

Home / Guides / EWMA Training Load

EWMA Training Load Explained: A Smoother Way to Read Your Workload

Q: What is EWMA in training load?

An exponentially weighted moving average (EWMA) blends every day's training load into a running value, weighting recent days most and letting older days fade gradually instead of dropping off a hard 7-day window edge. It models how fitness and fatigue actually decay.

Q: How is EWMA different from ACWR?

Classic ACWR uses block averages (last 7 days vs last 28 days), which have a cliff: a session counts fully one day and not at all the next. An EWMA ratio uses decaying weights so old sessions fade smoothly. Ægir Iron shows both side by side and flags when they diverge.

A plain weekly average forgets a hard session the moment it slides out of the window. An exponentially weighted moving average (EWMA) lets it fade instead, which is closer to how your body actually carries fatigue. Here's what that means and how to read the two numbers together.

The problem with a plain average

Most load monitoring, including the classic acute-to-chronic workload ratio, is built on block averages: your last 7 days, your last 28 days. Block averages are easy to compute and easy to explain, but they have a cliff built in. A brutal session from 8 days ago counts fully one day and not at all the next. Nothing about your body changed overnight; the math just dropped it.

Fitness and fatigue don't work like that. They decay gradually. A monster week still weighs on you two weeks later, just less than it did.

What an EWMA does instead

An exponentially weighted moving average blends every day's load into a running value, weighting recent days most and letting older days fade smoothly:

EWMA today = (today's load × λ) + (yesterday's EWMA × (1 − λ))
λ = 2 ÷ (N + 1), where N is the time constant in days

With a 7-day time constant (λ ≈ 0.25) you get an "acute" read that responds fast; with a 28-day constant (λ ≈ 0.07) you get a "chronic" read that moves slowly. Divide one by the other and you have an EWMA version of the workload ratio. The sports science case for this was made by Williams and colleagues (Br J Sports Med, 2017): decaying weights model how fitness and fatigue actually behave better than hard window edges do.

How the two reads differ in practice

A one-week spike: the rolling 7-day ratio jumps hard, because the whole spike lands in one window. The EWMA rises, but less, because it still remembers your quieter weeks.
A layoff: the rolling ratio falls off a cliff after day 7. The EWMA glides down, which is closer to how detraining actually feels.
A slow ramp: week to week the rolling ratio can look normal, while the EWMA quietly climbs. The creep is real; the 7-day window just can't see it.

Reconciling them: what a divergence means

Ægir Iron computes both from the same logged sets and shows them side by side. When they agree, either number tells the story. When they diverge by more than 0.25, the app flags it and says which situation you're in:

Rolling above smoothed: your last 7 days spiked against your longer-run trend. Expected during a planned peak or test week; otherwise, ease back in.
Smoothed above a low rolling ratio: you just backed off after a heavy stretch. If that's a deload, it's registering exactly as it should.
Smoothed above a normal rolling ratio: load has been climbing for weeks even though this week looks ordinary. The longer trend is the one to respect.

The 0.25 threshold is a design choice, not a law of physics. There is no published cutoff for when these two reads "meaningfully" disagree, so we picked a sensible default, disclose it, and stay quiet when your training history is too thin to trust either number.

The same honest caveat as ACWR

Smoothing the math does not rescue the injury-prediction claims that made the workload ratio famous; that research has been seriously challenged (see our ACWR guide for the full story). An EWMA ratio is a cleaner description of how your recent loading compares with what you've built up to. That description is genuinely useful for pacing a return from a layoff, catching an accidental ramp, and confirming a deload deloaded. It is not a crystal ball.

Ægir Iron tracks both reads automatically from your logged sessions, so the comparison is a by-product of training normally rather than a spreadsheet project.

Common questions about EWMA training load

What is EWMA in training load? An exponentially weighted moving average blends every day's load into a running value, weighting recent days most and letting older days fade gradually instead of dropping off a hard 7-day window edge.

How is EWMA different from ACWR? Classic ACWR uses block averages (last 7 vs last 28 days), which have a cliff. An EWMA ratio uses decaying weights so old sessions fade smoothly. Ægir Iron shows both side by side and flags when they diverge by more than 0.25.

What time constants does it use? A 7-day constant (λ ≈ 0.25) gives a fast acute read; a 28-day constant (λ ≈ 0.07) gives a slow chronic read. Dividing one by the other gives the EWMA workload ratio (Williams et al., 2017).

Does it predict injury? No. Smoothing the math doesn't rescue the injury-prediction claims, which have been seriously challenged. It's a cleaner description of your recent loading, not a crystal ball.

Track your load free →