Subscribing to a trading algorithm means buying a number. Usually a return: X percent a year, Y percent more than the market. The number is almost always calculated correctly. It just doesn't answer the question the buyer actually has: would I have sat through the path to it?

In German-speaking markets this business has three floors. On the bottom sits outright fraud: since June 2025 Germany's financial regulator BaFin has warned against a series of more than 700 near-identical websites advertising "automated trading of financial instruments by means of artificial intelligence from an entry investment of 250 euros" — no legal notice, no licence; the list was last extended on 6 August 2026, and 2026 brought further series of 48 and 89 domains. That floor takes thirty seconds to spot, so it isn't the subject here.

The top floor is more interesting: providers who work seriously, publish their method, and still leave out figures without which a strategy can't be judged. The five questions below were built from three of them.

Question 1: How far down did it go, and for how long?

The most detailed factsheet among the three providers checked here belongs to boerse.de-Signale Aktien Welt, published by TM Börsenverlag AG. Behind it sits the BOTSI Advisor, developed by Prof. Hubert Dichtl and Thomas Müller. It is methodologically serious: four trend and momentum indicators, each cited to the literature — relative strength per Levy (1967), cross-sectional momentum per Jegadeesh/Titman (1993), the 200-day moving average, and time series momentum per Moskowitz/Ooi/Pedersen (2012). Maximum ten positions, weekly rebalancing, daily risk monitoring.

As of 14 September 2026, the factsheet reports:

Hit rate, best trade, worst trade, average gain, plus 23 of 27 investment periods positive. All there. Everything except the figure that matters if you actually intend to follow the strategy: how far the portfolio fell in its worst stretch, and how long it took to get back.

The worst trade at −58.97 % is not the same thing. It describes one position, not the portfolio. With ten positions and a 44 percent hit rate, several losers can run at once, and that overlap is exactly the experience you have to sit through.

The American provider Arch Public looks the same. Its site shows seven test windows across three stocks, with alpha against buy and hold ranging from 20 to 524 percentage points and a top figure of +1,192.4 % for Nvidia on the weekly chart from January 2022 to June 2026 (buy and hold over the same window: +668.8 %). Drawdown appears nowhere. Neither depth nor duration, and no trade count or hit rate either.

Of the two figures, duration is usually the harder one. A 35 percent fall is survivable if it's over in four months. The same 35 percent spread across two years costs most investors their position — and usually at the bottom, because the patience ran out before the recovery arrived. Knowing only the return means not knowing what it cost.

Question 2: Backtest or live — and are they shown separately?

The boerse.de headline reads +27.1 % a year since 31 December 1999. Below every chart sits the note that all figures come from a historical backtest running from 31 December 1999 to publication on 29 April 2020.

That is stated openly, and it is the most important line on the page. Translated: roughly twenty years of the record are backtest, roughly six are live. Both sit inside a single number.

A backtest is built knowing what happened next. Live trading isn't. How the strategy has done since April 2020 is therefore the genuinely interesting information — and the headline doesn't contain it.

At Arch Public the situation is clearer and narrower at once: everything is backtest, as the fine print says ("hypothetical"). The seven windows have different start and end dates and different timeframes, which the provider itself flags with a note that the rows aren't directly comparable. No out-of-sample test or walk-forward is mentioned anywhere.

So ask for two curves instead of one. Where the backtest ends and live trading begins belongs in the chart, not in a footnote.

Question 3: Which costs are included — and does that include the subscription?

boerse.de states its assumptions: 0.1 percent in fees per transaction, dividends and taxes excluded, calculated in euros, each new position opened at ten percent of portfolio value. Worked through: 84 trades a year are around 170 buys and sells, each on a tenth of the portfolio, each charged 0.1 percent — about 1.7 percent of the portfolio a year in assumed fees. The assumption is not a detail: a broker charging 0.25 percent costs another two and a half percentage points a year against the published calculation.

What no return figure includes is the subscription itself. The annual subscription costs €1,019.20 (a promotional price with 25 percent off; the regular price is around €1,360) and the recommended minimum capital is €20,000. That is a little over five percent of deployed capital a year before a single trade runs, close to seven at the regular price. Against a reported 27.1 percent there is plenty of room. Against a weaker year there isn't.

At Arch Public the strategy runs as an invite-only script inside the user's own TradingView account. The commission in the backtest is a field the user fills in; the documentation notes that exchange fees are tiered and change over time. For a strategy that reacts to individual candles on the one-hour and four-hour chart, that line item is what decides the alpha.

The calculation you have to do yourself: subscription fee divided by your actual capital. That is the return the strategy has to earn before anything reaches you.

Question 4: Which benchmark — and which version of it?

boerse.de compares against the MSCI World and reports a gain of 195.2 percent since the end of 1999, or 4.1 percent a year. That figure matches the price index, which ignores dividends. The net return index in euros, the one ordinary ETFs track, has run at roughly 6 percent a year over 25 years according to third-party analyses of MSCI data — cumulatively closer to 370 percent than to 195.

Formally the choice is consistent: the strategy also excludes dividends, so both sides exclude them. But the same rule lands very differently on each side. The strategy holds a stock for 33.9 days on average and collects barely any dividends anyway. A buy-and-hold comparison collects all of them. Dropping them costs one side almost nothing and the other around two percentage points a year.

Applied to the reported outperformance of 23.0 percentage points annually, about 21 would remain against the net index. The finding doesn't flip. But it moves measurably, purely through the choice of an index variant.

So the question is never just which index. It is which version of it, in which currency, over which window.

Question 5: Who produced the figures — the provider or a third party?

Algo-Camp of Berlin sells Expert Advisors for MetaTrader, with one-off licences between €397 and €7,020 instead of subscriptions, German-language support and a registered address. A detailed review on kagels-trading.de (as of 7 September 2026) lists as a weakness, verbatim: "The performance claims come from the provider; a public live-money account is missing."

That is not a small thing, and it isn't an accusation against particular firms. It is the industry's normal state: the number used in the advertising comes from whoever sells the product.

What a third party can establish is limited but not nothing. A publicly visible live account. A track record maintained by an independent body. Or a strategy described completely enough that you can recompute it yourself.

The last of those is the strongest, and boerse.de comes closest to it: all four indicators are named with sources, the portfolio rules are published, the cost assumption is published. That is enough to rebuild the backtest in principle. Such openness is the exception in this market and deserves saying out loud.

What to do with this

Five questions, in order:

1. How deep was the maximum drawdown, and how long did it last?
2. Where does the backtest end and live trading begin?
3. Which costs are included, and what percentage of my capital is the subscription?
4. Which benchmark, which index variant, which currency?
5. Who generated the figures, and can a third party verify them?

None of these imputes intent to anyone. They test whether a return carries enough context to mean anything.

Underneath all five sits the same thing. A return is an average. It says nothing about the distribution behind it — not the spread of the trades, not the depth of the setbacks, not their duration. Knowing only the average means not knowing whether you would have survived the path to it. That is the figure that belongs first, not last. In the Backtesting Arena it therefore sits in every result right next to the return: maximum drawdown, hit rate, trade count and the buy-and-hold comparison over the same window.

---

Not investment advice, not a recommendation, not a forecast — historical patterns are no promise. All provider details are taken from their publicly accessible pages as of 14/15 September 2026.

Sources: TM Börsenverlag AG, boerse.de-Signale Aktien Welt, product page at boersenverlag.de (as of 14 September 2026) · Arch Public, archpublic.com and docs.archpublic.com (15 September 2026) · Karsten Kagels, "Algo-Camp Erfahrungen", kagels-trading.de (as of 7 September 2026) · BaFin, consumer notice "Investieren mit KI" (3 June 2025, updated 6 August 2026) and platform-series warnings of 13 February and 13 May 2026 · MSCI World Net EUR, 25-year return per investingintheweb.com based on MSCI data

Study the Past — Improve your Future 🥋