What does a backtest say about the future?

Only how a signal would have worked in the past. It is an indication, not a promise, and the weaker the fewer cases and episodes stand behind it.

Why do you report weak results?

Because a backtest is only credible if you also show where it adds nothing. Tesla and Porsche SE are good examples.

How well have your fair values done so far?

It is too early to say. The oldest analyses date from Sept 20, 2026; the first three-month evaluation is possible from Dec 20, 2026 at the earliest. We will add it here then.

Knowledge · Test it

Backtesting a valuation: how reliable is a hit rate?

A backtest checks how well a signal or model would have worked in the past. You always compare with a base rate, the result without the signal.

At a glance

Our method: We find all days on which a condition held (e.g. “below SMA200 and RSI oversold”) and measure the hit rate after 5, 10 and 20 days against the base rate of all days.
What matters is n (sample size) and the number of independent episodes. Many days from few episodes suggest a certainty that does not exist.
We also publish weak or contradictory results.

Formula

Edge = hit rate with signal − base rate   (in percentage points)

Worked examples

Stock	Condition / signal	Result	Base rate	n	Assessment
Alphabet	below SMA200 + RSI oversold	82 / 76 / 70%	57 / 58 / 61%	79 days, few episodes	anecdotal
Merck	above SMA200, RSI neutral	+0.6 / +1.3 / +1.9 pts	–	9,932 days	small but robust
Tesla	Daily MACD bull cross	+0.6 / −0.7 / +1.5 pts	–	51	no edge
Porsche SE	Death cross daily vs. regime weekly	−2.94% (20 days) vs. +10.75% (52 weeks)	+0.46%	approx. 10 independent windows	contradictory
SpaceX	–	deliberately no backtest	–	history since IPO too short	–

Figures apply to 5 / 10 / 20 days. The chart shows the edge over the base rate. Porsche SE is missing there because the measures are not comparable.

Backtest: edge of the signal by example

What the number does

The most impressive figure (Alphabet, 82%) has the thinnest basis.
The least conspicuous (Merck, +1.9 points) is the most reliable.
That is exactly what this page should convey: a rate without sample size and base rate is worthless.

Pitfalls

Overlapping windows: Consecutive days are not independent.
Few episodes: 79 days can consist of a handful of market phases.
Overfitting: If you test many signals, you will always find one that looks good.
Regime change: The market no longer behaves as it used to.

FAQ

What does a backtest say about the future? Only how a signal would have worked in the past. It is an indication, not a promise, and the weaker the fewer cases and episodes stand behind it.
Why do you report weak results? Because a backtest is only credible if you also show where it adds nothing. Tesla and Porsche SE are good examples.
How well have your fair values done so far? It is too early to say. The oldest analyses date from Sept 20, 2026; the first three-month evaluation is possible from Dec 20, 2026 at the earliest. We will add it here then.