Timeseries Finance

Time series for finance

Financial time series carry traps that general charting advice does not cover: periods are discrete and closed, values get restated, definitions change mid-series, and the thing people want to compare usually has a different magnitude than the thing beside it.

Discrete periods are not continuous

Monthly data is monthly. A line smoothed through monthly points invents intermediate values that never existed, and a reader will read them off.

• Columns for discrete periods where each period is a closed fact — monthly revenue, quarterly bookings.
• Lines for continuously-varying quantities sampled over time — cash balance, headcount, open AR.
• Never spline/smooth between periods. Straight segments, or columns.
• Break the line at missing periods. chartkit.linePath does this by default, deliberately — connecting across a gap draws data you do not have.
• Mark partial periods (month-to-date) distinctly. An incomplete month plotted alongside complete ones reads as a collapse.

Instead of dual axes

Dual axes create a crossover point that is an artifact of two arbitrary scales, and readers infer meaning from it. Three honest substitutes:

1. Index both to 100 at a common base period. Now they share a scale and the comparison is real.

index_t = 100 × value_t / value_base

Best for comparing growth rates of differently-sized things. Label the base period in the axis title — an index chart with no stated base is unreadable.

2. Two stacked panels sharing an x-axis. Costs vertical space, keeps both scales honest, and aligned time makes the relationship visible anyway.

3. Chart the derived measure. If the reason for two axes is "revenue and margin", chart margin % alone. Usually the second axis exists because the real question was a ratio.

Log scale — when and how

Use a log y-axis when the series spans more than roughly two orders of magnitude, or when the question is about growth rate rather than absolute change. On a log scale, equal slopes mean equal percentage growth, which is exactly what a growth-stage conversation is about.

• Label it clearly — most readers will not notice otherwise, and a log chart read as linear is badly misread.
• Ticks at powers of ten with intermediate gridlines.
• Log cannot show zero or negatives. A log chart of a series that crosses zero is not possible; use linear.
• Do not reach for it by default. Most finance series are fine on linear, and log costs interpretability.

TTM, run-rate, and point-in-time

Three different series, routinely confused. Label which one is plotted.

Never mix them on one chart without explicit labelling. A TTM line beside a point-in-time line looks like a divergence when it is a definitional artifact.

For burn and runway specifically, use a trailing 3-month average rather than the latest month — a single month's burn is not a rate.

Seasonality

• YoY comparison is the honest way to handle it: same month, prior year. Seasonality cancels.
• Overlay prior-year as a muted line on the same axis for a same-period comparison.
• Small multiples by year — one panel per year, months on the x-axis — makes the seasonal shape visible directly.
• Do not deseasonalize a short series. With under about three years of history the adjustment is fitting noise.

Forecast boundaries

Where actuals end and projection begins must be unmistakable:

• Solid line for actuals, dashed for forecast, with a vertical rule at the boundary labelled "forecast →".
• A single forecast line understates uncertainty. Prefer a band. See uncertainty-charts.
• Label the forecast vintage: "Forecast (Aug-26 reforecast)". Forecasts get revised, and an unlabelled one is unreconcilable.
• Never continue a confidence-inspiring solid line past the last actual.

Annotation

Events explain shape. A chart without them makes the reader ask questions the chart could have answered.

• Vertical rules with short labels for discrete events: a price change, a launch, an outage, a funding round, a definition change.
• Shaded regions for periods: a pricing pilot, a hiring freeze, a COVID-like disruption.
• Direct callouts on the two or three points that matter — a peak, a trough, the current value.

A definition change gets an annotation and a visible break in the series. A smooth line across a definition change is a false continuity claim. See chart-annotation.

Alignment and axis discipline

• Zero baseline for additive measures — revenue, cash, ARR, headcount. Truncating doubles the apparent size of a change.
• Truncation is fine for rates and ratios where zero is not meaningful; say so in the axis label.
• Consistent period boundaries. A chart mixing calendar and fiscal quarters is wrong even if every point is right.
• Date labels: Jan 26, Q1 26, 2026 — never bare 1, 2, 3.
• Time runs left to right. Always. This should not need saying, and yet.

Restatement

Prior periods change — a reclass, a correction, a re-segmentation. When they do:

• Do not silently redraw history. A chart that quietly moves last quarter destroys trust in every number on the page.
• Annotate the restatement and, where it matters, show both the as-reported and the restated series.
• State the as-of of the data (microcopy-and-states), so two versions of the same chart can be reconciled later.

Related skills

• chart-selection — line versus column versus something else
• uncertainty-charts — forecast bands
• variance-charts — actual against plan over time
• svg-charting — linePath and its null handling
• chart-annotation — events, vintages, and definition breaks