Data and method

How candidates are picked

Ranked from the lowest EV/EBIT: what it would cost to buy the whole company including its debt (enterprise value = market value + debt − cash), as a multiple of its operating profit. Lower is cheaper. It is like a P/E ratio, but counts debt and cash and uses operating profit, which interest and one-off items distort less. Only companies with a Piotroski F-score of 7 or more out of 9. The F-score counts nine checks: profitable, positive cash flow, and profits, balance sheet and efficiency improving on the year before. In the past tests, this was the only method compared that beat the market in all four (two periods each in Japan and the US). Return on capital (Greenblatt's definition: operating profit ÷ (net working capital + net fixed assets)) and warning signs are shown for reference. For ranking only, enterprise value is not taken below 30% of market value, so companies whose cash nearly equals their market value do not show absurd yields.

How the method was chosen

We compared five methods on figures known at the time, in two periods each for Japan and the US: cheapness alone, Greenblatt's magic formula, each with an F-score filter, and cheapness with a return-on-capital floor. Cheapness + F-score 7+ was the only one to beat the market in all four. We picked it from the five, so the past results flatter it somewhat.

Warning signs

Loss-making, shrinking results, profit not backed by cash, heavy debt, share dilution, and possible one-off profit are shown for reference only. The F-score already covers most of them.

AI

For each candidate, AI (Claude, via the operator's subscription) is given the figures and excerpts of the latest annual report or 10-K, and drafts hypotheses on why the stock is cheap, what could close the gap, and how a buyer could still lose. It does not search the web. Each point must quote the filing; quotes are checked by machine and unmatched points are dropped. AI runs only when a company newly becomes a candidate, files a new annual report, or 90 days have passed; at most 15 companies a day.

Sources

Japan: EDINET annual reports, Yahoo Finance prices. US: SEC EDGAR XBRL (frames and company facts) and 10-K text, Yahoo Finance prices.

Limits

Two test periods per market; both Japanese periods fall in years that favoured cheap Japanese stocks. Japanese figures come from the annual report, so they can be up to a year old. Some companies that disappeared could not be followed in the tests. AI hypotheses are drafts, not research.

Operator

After Deal is run by an individual as a hobby, unconnected with any employer. Not investment advice.