Built a composite of Graham/Buffett/Greenblatt/Lynch + a Damodaran DCF over ~2,400 names. Fair values run below market almost everywhere. Roast the model, not the market.
Been building my own valuation stack on the side for about a year, all free data (SEC filings, mostly). Not selling anything, everything on the site is open and you can pull the whole [evaluations CSV](https://kabrios.com/data/evaluations.csv) without signing up. Looking for people who do this for a living to tell me where the methodology is dumb.
Short version of how it works ([full writeup here](https://kabrios.com/methodology.html)): two separate verdicts. A 1-5 business quality composite, and a price zone that compares the market price to a fair value with a margin of safety that widens as uncertainty goes up. The composite is four pillars: Quality 38% (Buffett ROE consistency + Greenblatt return on tangible capital), Growth 23% (Lynch-style), Financial Strength 23% (Graham's balance sheet tests), Trend 15% (O'Neil / relative strength). The [DCF](https://kabrios.com/damodaran.html) is deliberately 0% in the composite and only feeds the price side. Every metric scores continuously between four published anchors instead of buckets, and the anchors are absolute, not percentile, so how many names get a given rating floats with the market.
The problem. Of the 590 names where the system is willing to publish a fair value at all (the other \~1,800 it refuses, mostly for data quality reasons), the median price/fair is 1.05x, 20% of them are above 1.5x, and the ones that trade above fair are basically all the names everyone loves. [AAPL](https://kabrios.com/AAPL.html) scores 4.4/5 on quality and comes out at $210 fair vs $330 market. [GOOGL](https://kabrios.com/GOOGL.html) is 4.8/5 and $208 vs $338. Meanwhile [NFLX](https://kabrios.com/NFLX.html) shows up at 0.78x. Only 4% of the full universe lands in a buy zone.
So either the market is rich, or I've built a machine that's structurally bearish, and I honestly can't tell which. My suspicions, in order:
1. I'm using a stale/too-high equity risk premium and a conservative terminal growth cap, and the two compound. Everything I did for "prudence" pushes the same direction.
2. My fair value is probably 70%+ terminal value, which means it's really one assumption wearing a ten year model.
3. Publishing a point fair value at all is the wrong output. I keep thinking I should run it backwards instead, solve for what growth / margin / duration the current price implies and just say whether that's plausible.
On the composite side, the thing that bugs me most is that Graham's tests were pass/fail and I turned them into a continuous score that averages in. So a strong growth number can paper over a failed balance sheet test. Also Quality is 38% but it's ROE + ROIC, which are cousins, so the real loading on "returns on capital" is higher than the stated weight.
What I'd actually like answered by someone with a desk:
* Is reverse DCF (price -> implied assumptions) how you'd handle this, or is that a cop-out?
* Which ERP do you use, trailing or implied? Is that where my gap is coming from?
* Do you keep hard vetoes in a composite (fail the balance sheet, capped at X regardless), or is additive fine?
* In a market you'd call normal, what % of your coverage sits materially below your fair value? If the answer is 10-15% my bands are too tight. If it's 3-5% then maybe it's just expensive and I should stop fighting it.
I started a [calibration page](https://kabrios.com/calibration.html) tracking every call in an append-only ledger so in a couple years I'll know if the buy zones beat the sell zones. Right now it's 5 days of data so it's noise. Wanted to put the model in front of people before I have years of being wrong.
Happy to answer anything about the data pipeline too.