When does Misano use the median instead of the latest year?

When residual coefficient of variation exceeds 0.30, or when there is a profit↔loss sign flip in the last two reported years. The method label says normalised (median) — it is not silent.

Do old startup losses force a median forever?

No. Older losses further back than the last two reported years do not brand the company forever. The sign-flip rule looks at recent switches only.

Is this the same as adjusting owner pay and one-offs?

No. Buyer-restatement (owner compensation, one-offs, related-party contracts) is a different layer, covered at misano.ai/guides/how-to-value-a-private-company. Trend-conditional normalization decides which year-level enters the multiple when the series is noisy.

Growth is not volatility.

Q: What does trend-conditional earnings normalization mean in Misano?

Misano fits an OLS trend through the EBITDA (or net profit) series and measures residual noise as RMSE of residuals ÷ |mean|. Above 0.30, the median of available years goes into the multiple with a normalised (median) label. Steady growth is a trend, not volatility. Full method: misano.ai/guides/how-we-value-companies-at-scale.

One cyclical peak × a multiple is a lottery. Misano's earnings normalization is trend-conditional: it fits an OLS trend through the EBITDA (or net profit) series and measures how noisy the residuals are. Steady growth is a trend, not volatility. When residual noise crosses a threshold, the median of available years goes into the multiple — and the method label says so.

How the test runs

Fit an OLS trend through the EBITDA series (or net profit when that is the metric in play).
Coefficient of variation of the residuals = RMSE of residuals ÷ |mean|. A steadily growing company has residuals near zero.
Above a threshold of 0.30, the latest year is treated as untrustworthy for the multiple → the median of available years is used, with the label normalised (median).
Sign-flip rule: a profit↔loss switch in the last two reported years triggers automatic median — even if the residual CV alone would not.

What it deliberately ignores

Older losses (startup history further back than the last two reported years) do not brand the company forever. The sign-flip rule is about recent switches, not ancient red years.
This is not the buyer-restatement layer (owner compensation, one-offs, related-party contracts) described at misano.ai/guides/how-to-value-a-private-company. Trend-conditional normalization decides which year-level goes into the multiple when the series is noisy.
It is not a rewrite of staleness gates or the method cascade. Hard-stale metrics still cannot headline; EV/EBITDA still prefers usable normalized EBITDA. See misano.ai/guides/stale-filings-cannot-make-todays-headline and misano.ai/guides/ev-ebitda-then-pe-then-ev-revenue.

Why the label matters

The method label is a sentence, not a code — e.g. EV/EBITDA 12× (manual override; normalised (median)). Screening without that provenance is how a peak year becomes a fake-precise EV.
Normalization feeds the same multiples band that triangulates with Damodaran DCF and NAV on the football field: misano.ai/guides/value-band-is-triangulated.

What this is not

Not silent averaging of every year into a fake-smooth earnings path without a threshold test.
Not carrying last year's EBITDA forward when this year's is missing — that is a cascade drop, not normalization.
Not a rewrite of the full valuation essay. Multiples cascade, staleness, NAV, DCF, and football field stay at misano.ai/guides/how-we-value-companies-at-scale.

Get started at https://app.misano.ai/login. From €39/month billed annually, 14-day money-back, no sales call. Product path: misano.ai/scout.

See also: misano.ai/guides/how-we-value-companies-at-scale, misano.ai/guides/ev-ebitda-then-pe-then-ev-revenue, misano.ai/guides/stale-filings-cannot-make-todays-headline, misano.ai/guides/value-band-is-triangulated, and misano.ai/guides/how-to-value-a-private-company.