How Leeway evaluates stocks

Context: A stock hypothesis does not start with the stock, it starts with the market. Inputs are the current meta market hypothesis, the three ratings (Business-Rating, Market-Fit-Rating, Cycle-Rating), fundamentals, price history and the peer group. The ratings say what the company is. The hypothesis asks what actually sets its price.

Analyze: Two independent language models work the same question separately, without seeing each other's answer. Each builds its own mental model of the company, names for itself which facts it needs verified, and formulates its own research questions. Those questions go into source-based research; the answers come back with URLs, not from model memory.

Reasoning: Every initial hypothesis is attacked before anything is built on it:

• Devil's advocate: hunts for the point where the thesis breaks: hard factual errors and broken causal chains, not style. A discussion round then works through the critique.
• Premise check: tests whether the load-bearing assumptions are actually covered by the researched sources. What is uncovered gets flagged as uncovered instead of quietly carried forward.
• Valuation critique: does the counter-math: what owner earnings would actually have to arrive over the coming years for today's price to work out? With its own source research attached.
• Market-reaction critique: checks the thesis against how the stock really trades: volatility, reaction patterns to news, who is actually trading it.

Only then are the two strands merged, and the merged version runs through the same critique again.

Conclusion: The output is not a buy recommendation but a testable thesis: the mechanism that sets the price, scenarios instead of a single target, friction points and catalysts, and a list of what would falsify it. Every source and every AI call is stored with the document, so any run can be traced backwards. And a thesis like “this company is not an earnings compounder but a levered spread option on three price differentials” can be checked and refuted. That is the whole point.

Context: The CIO Perspective is not a fourth rating and does not feed into Leeway-Score. It answers a different question: which regime is the market in right now, and which factor is being paid for in it? The basis is the meta market hypothesis.

Analyze: Inputs are around ten continuously maintained driver assessments, the price context of the major benchmarks (US breadth, small caps, Europe, emerging markets, bonds, gold, currencies) and data from outside the price series. Here too, two independent models work separately. Then comes a dedicated pass with a single question: which context is missing to judge this correctly? Those gaps are researched deliberately, together with a factual timeline of the last 6 to 24 months with sources.

Reasoning: Two steps make the difference:

• Actor model: who actually moves these prices, and under what constraint? Not “the market”, but concrete actors with concrete obligations: insurers with guaranteed rates, passive inflows, government financing, levered positioning. Prices come from constraints, not opinions.
• Move check: has the move already happened? Every proposed overweight and underweight is held against the actual price development of recent months. A correct thesis the market has already priced in is no longer a position.

Only then are the strands merged and the allocation sharpened.

Conclusion: The result is a judgment about the market, not about a stock: the regime in one sentence, overweights and underweights with reasoning, several scenarios with probabilities instead of a single forecast, and a handful of tickers as observation posts. The three ratings stay company-level. The CIO Perspective says whether the market is paying for that factor now. A strong Business-Rating in a regime that does not reward quality is not a buy case. Market environment first, then the company.

What we believe

Institutional capital flows leave traces in the market before they show up in the media. Read those traces and you no longer need the headline. That is why Leeway deliberately shows few, decision-relevant metrics instead of a news stream. Information overload does not improve investment decisions, it degrades them. And to us a share is an ownership stake with voting rights, not a ticker on a chart. The long version is under About us.

It also explains why Business-Rating comes from a language model rather than an analyst: founder Lars Wißler is a computer scientist whose first thesis dealt with automated text processing in the context of algorithmic stock analysis, roughly 15 years before the current AI hype. The advisory board includes Sven Krause, a former head of fund management with 30 years of capital-market experience and a focus on quantamental investing.

What counts as proof here

Every Leeway rating is, at its core, a backtest claim, and backtests are cheap to fake: with enough data and sloppy methodology you can manufacture almost any result you want. The Whitepaper therefore states openly what we accept as evidence:

• Sample size: patterns with fewer than 100 hits are discarded. Below thirty hits any claim whatsoever can be produced; from around 300 a sample becomes reliable, from 1,000 it is considered representative. “Predicted every bear market of the last 20 years” is two hits, which is nothing.
• System evolution: markets are not a constant of nature. Data from 50 years ago describes a different system. Scoring therefore runs on the last three years, and the most recent 250 trading days serve as the test set so that changed market conditions become visible instead of being papered over.
• Cross-validation: the data is split into several groups, each with its own training and test set. Whatever works in only one group is chance.
• Randomisation: random portfolio walks execute only a randomly chosen share of the signals, like a real user who does not trade every signal. Whatever only works under perfect execution does not work.

The limit belongs with it: no backtest guarantees the future. We document performance, risk metrics and the boundaries of the method. That is precisely why Business-Rating, Market-Fit-Rating and Cycle-Rating stand side by side instead of one single number claiming to be the answer. The full derivation of Market-Fit-Rating is on its methodology page.

What the number is built for

Leeway-Score is optimised for a one-to-two-year horizon and for stable, low-volatility price development: solid long-distance runners, not the next ten-bagger. Strategies shows how that turns into a portfolio: sort via Rank & Screen, usually by Market-Fit-Rating, then watch diversification: no country and no sector above 20–25% weight, at least eight countries and sectors.

What the score explicitly does not do: decide your cash quota and your entry timing. That is a separate layer made of market timing and the CIO Perspective. A high Leeway-Score tells you which company. It does not tell you how invested you should be right now.