AKM1 — The Controversial Model: SUPER DEEP

AKM1 in one paragraph: the model scores a company on 20 variables divided across seven categories — Growth (V01–V03), Valuation (V04–V06), Profitability (V07–V09), Stability (V10–V12), Moat (V13–V15), Catalyst (V16–V18) and Risk (V19–V20). Each variable earns 0, 1, 2, 3, 4 or 5 points against documented thresholds. The arithmetic is simple: 20 variables times a maximum of 5 points gives exactly 100 points as the maximum total.

Six categories have three variables and contribute at most 15 points each; Risk has two variables and contributes at most 10. The score is a fundamental profile — a wave of evidence — and never a buy signal in itself. That is where the first controversy begins, because according to the academic mainstream, expected returns should be explained by factors, not by headings with points in them.

The 0–5 scale is a deliberate compromise. A ten-point scale would give an appearance of precision we do not have — no one can honestly tell a 7 from an 8 on brand strength. A three-step scale (bad–okay–good) is not enough to separate a marginal winner from an ordinary player. Five steps force you to take a position without demanding false accuracy. The rule we work by: every point must be justifiable in a single sentence with a number or a documented fact in it.

The thresholds you meet in this course are guideline values — sector adjustment is permitted and encouraged, but it must be documented, not smuggled in. Scoring without justification is not analysis; it is a mood with numbers attached.

Margin of safety is the model's commander-in-chief — and it deliberately has no V number. Benjamin Graham devoted chapter 20 of The Intelligent Investor to the principle: the difference between price and intrinsic value is what protects you when your analysis, or the world, is wrong. AKM1 scores the company's fundamental data; the margin prices everything. A straight 100-point score on a company trading at four times intrinsic value is a weak case, not a strong one.

Why no variable? Because margin of safety is not a property of the company — it is a property of the transaction. Baking it into the total would let you say that an expensive company with 92 points is better than a cheap one with 78. That is exactly the wrong order. First the score, then the price. Always.

The map of the controversies has three fronts, and you should know all of them by heart. Front one: the moat variables V13–V15 (patents, brand, network effects) — a quantitative tradition rooted in Fama's efficient market hypothesis (1970) would call them unrecordable opinion data that cannot be backtested. Front two: the catalyst variables V16–V18, which score future events — mainstream models eat only backward-looking data.

Front three: the weighting matrix itself, where no regression weights have set the numbers. Against all of this stands an honest confession the course repeats: AKM1 is a pedagogical structuring tool that forces you to look at everything — not a proven source of alpha. We believe in the structure. We do not have a p-value to show for it. That is the starting position, and it is better to know it before your opponent does.

The matrix in concrete terms: Growth (V01 Revenue growth, V02 ARR growth, V03 Revenue diversification) max 15 points. Valuation (V04 P/S, V05 P/B, V06 EV/EBITDA) max 15. Profitability (V07 Gross margin, V08 EBITDA margin, V09 ROE) max 15. Stability (V10 Debt-to-equity ratio, V11 Liquidity, V12 Revenue stability) max 15. Moat (V13 Patents and IP, V14 Brand and customer loyalty, V15 Network effects) max 15. Catalyst (V16 Product launches, V17 Agreements and partnerships, V18 Regulatory catalysts) max 15.

Risk (V19 Capital burn and share issue risk, V20 Share buyback) max 10. Total: 100. Notice what the matrix does not do: it does not multiply, it does not discount, it has no hidden interaction terms. A 5 in moat counts exactly as much as a 5 in gross margin. That is simple, transparent — and attackable, which we will get to.

The first objection we dismiss ourselves: why not equal weighting between the categories, seven categories at roughly 14,3 points each? Three reasons. One: the variable — not the category — is the atomic unit of judgement. It is possible to justify a threshold for gross margin; it is not possible to justify why Profitability as a box should weigh exactly as much as Catalyst as a box. Two: category weighting forces unnatural variable weights.

If Risk were to give 14,3 points with two variables, V19 and V20 would weigh 7,15 each — more than any single variable in the model — or we would have to invent a third risk variable just for the sake of symmetry. Three: the role of the Risk category is largely asymmetric — it flags and downgrades, it does not build stories. That it contributes a ceiling of 10 is a conservative design, not an oversight.

What mainstream does — and here we must be precise, because it is a strong opponent. The Fama-French factor models (three factors in 1993, five factors in 2015) are long-short portfolios: HML is long high book-to-price and short low; SMB is long small and short large; in the five-factor model RMW (long robust profitability, short weak) and CMA (long low investment, short high) are added. The weights there are not doctrinaire guesses — they are data-driven constructions tested over decades of cross-sections, and when Fama and French showed in 2015 that HML is largely swallowed up by the profitability and investment factors, they did exactly what an AKM1 critic would demand: they revised the model.

That is scientific behaviour, and we respect it. The difference is the purpose: a factor model seeks to explain return differences between portfolios. AKM1 seeks to structure an overall assessment of individual companies for educational purposes. We are not constructing a hedge fund strategy, and we will not pretend that we are.

The real weakness — and the honest confession: that every variable weighs equally (5 points at the ceiling) is itself a choice no regression has made. An optimizer allowed to choose weights freely would almost certainly twist the matrix — deciding that profitability should weigh more and the catalysts less. Why do we not let it? Because freely optimized weights on small data are a factory for overfitting: Bailey and López de Prado have shown how backtests with many degrees of freedom produce impressive Sharpe ratios that collapse out of sample.

Choosing between a falsely precise optimized matrix and a transparent, simply explicable matrix that each student can review in their head — we choose transparency. The counter-demands we place on ourselves: documented scores, sensitivity tests (drop an entire category and see whether the overall picture changes — if it does, the picture was fragile), and the courage to change thresholds publicly when they fail. A matrix that may not be reviewed is not a model; it is dogma.