Seeing Signal in a Time Series

Most mistakes with data come down to one of two errors: reacting to noise, or missing a signal. Get it wrong in one direction and you chase phantoms. Get it wrong in the other and you sleepwalk into trouble. This guide builds a chart design, layer by layer, that makes the distinction visible.

1. The Level

Plot the raw values. You see the broad arc — up, down, sideways — but not the dynamics. A line rising steadily at 5% per year looks identical to one that grew 8% last year and 2% this year. The level tells you where you are. It does not tell you what is happening.

2. Year-over-Year Change

Compare each observation to the same period twelve months earlier. This strips out seasonality and reveals what the level chart hides: acceleration, deceleration, turning points. Green above zero means expansion; red below means contraction.

Toggle between the two views. The level chart shows a smooth recovery from the 2020 shock; the YoY chart reveals a massive positive spike in early 2021, not because anything extraordinary was happening, but because the comparison base from 2020 was so depressed. This is a base effect: a distortion caused by an unusual prior period, not by current conditions.

3. The Trend

A single YoY reading is hard to judge. Is +5.2% good or bad? Fit a linear regression to the YoY values. The slope tells you whether growth is speeding up or winding down. Readings above the line are outperforming; readings below are underperforming. The line gives each data point a context it cannot have on its own.

But a single deviation from trend doesn't necessarily mean anything has changed. Some variation is routine. The question is how much variation is too much.

4. Deviation from Trend

Subtract the trend from each observation. What remains is the residual — how far reality has drifted from the trend. Draw limits at ±2 standard deviations. Points inside the limits are routine variation. Reacting to them almost always makes things worse. Points outside are signals: something has changed.

In this sample, 5 readings breach the limits, all concentrated around the 2020 shock and its aftermath. The readings in between, even those that look volatile in the YoY chart, fall within the limits.

The ±3σ threshold means fewer false alarms, at the cost of slower detection. The standard deviation is exponentially weighted: recent residuals count more than old ones, so the limits tighten when the series has been calm and widen when it has been volatile. A second rule catches subtler shifts: if enough consecutive points land on the same side of the center line, the process has likely moved even though no single point breached the limits.

5. The Write-Up

Each chart carries a sentence that says what the numbers say, so the reader does not have to decode the chart. Here is the template:

The trend value and slope come from the regression. The sigma reading comes from the residual. The base effect explains why the current YoY looks the way it does — when the comparison point from a year ago was unusually high, it mechanically drags the current reading down even if nothing has changed on the ground.

A reader scanning twenty series should be able to read the sentence and know whether to look closer or move on.

6. The Complete Picture

The card below puts all three panels together, with one addition: a dashed indigo line extending from the last YoY data point. This is the forward projection — where YoY would go mechanically over the next few months if the level stays flat, based purely on the comparison base rolling forward. It previews where the arithmetic is headed before the economy has its say.

The level shows where you are. The YoY shows how fast you are moving. The deviation shows whether the speed is unusual. The projection shows where the road bends next, even if you hold the wheel steady. In 2020, the level showed a V-shaped dip, the YoY chart revealed the dip followed by a base-effect spike, and the deviation chart flagged both as signals. Today, the series is near an all-time high, growing at a moderate rate, with deviations within normal bounds.

Every data point invites a reaction. The discipline is knowing which ones deserve one.