The Profitability Turn Was Not the Buy Signal

The first year a blitzer prints operating profit feels like the end of the war. We tested whether buying that moment paid. On our own pre-registered criterion, the answer is no — and the failure is the point.

5 min read · 18 events · 42 companies · Not investment advice

We locked the verdict before seeing the data

This study ran under a rule we intend to keep: the hypothesis, the exact event definition, the entry rule, and the falsification criterion were written into a pre-registration file (HYPOTHESES_PREREG_2026-10-13.md) before a single return was computed. Changing definitions after seeing results was prohibited — spots where a definition looked imperfect afterward are logged as "definition issues," not fixed retroactively.

The question on trial, labeled H2: is the first year of GAAP operating profitability a good entry point? The event was defined mechanically: the first fiscal year with operating income above zero following at least one loss year. Entry was set at the first trading day after that year's 10-K filing — so no look-ahead, no hindsight. Three controls: (a) buy-and-hold the same company from the first available price, (b) entering after the peak-growth year's 10-K, and (c) the S&P 500 over the same windows.

The rejection criterion was equally mechanical: if the median +3-year return of inflection entries is at or below the paired same-company buy-and-hold median, the hypothesis is rejected. No room to bargain afterward.

The method, in brief

We read only the sign of annual GAAP operating income across 42 companies with filed financials — no adjustments, no judgment calls. That sweep produced 18 inflection events, 9 always-profitable companies (no event possible), and 13 that have never inflected (a built-in control against survivorship optimism). Of the 18, 15 had computable returns: CORZ was delisted at its filing, CRWV was still private, and DASH's forward window hasn't arrived yet. Filing dates came from EDGAR submissions; prices are split-and-dividend-adjusted closes.

This test is the second of four hypotheses we registered together (H1 threshold-crossing, H3 score momentum, H4 valuation-zone revalidation). H2 went first because it had the cleanest data and the most complete event set. The individual events show how little any single story decides: Palantir, the lone giant, turned in 2023 and returned +345.7% in the year after its filing. AbCellera also "turned" in 2020 — on a one-time Lilly royalty that then collapsed — and lost 71.3% in a year and 89.7% over five. Tesla's turn (2020, entered February 2021) gave +9.7% then −31.6%. MercadoLibre (−30.5%, one year) and Shopify (−8.2%, one year) were the two large platforms in the 65-point-score combo, and neither rewarded the entry. Eighteen events, eighteen different markets, eighteen different verdicts — which is precisely why the aggregate, not the anecdote, carries the verdict.

The verdict is REJECT: the +3-year median for inflection entries (−8.3%) sits below the paired buy-and-hold median (−1.2%). The secondary criterion agrees — the median excess return over the S&P 500 at three years is −50.5 percentage points. Only one of nine entries beat SPY over three years.

The combination we most wanted to test — a Blitz Index score of 65 or higher held alongside the inflection — is undecidable: three observations (MercadoLibre −30.5%, Palantir +345.7%, Shopify −8.2% at one year). One enormous winner dominates the whole subsample. We report it as insufficient evidence, not as a conclusion.

Why it lost: the market saw it coming

The interpretation is straightforward, and it is not that profitability doesn't matter. A profitability inflection is a business milestone, not a price bottom. By the time the 10-K announces the turn, the expectation is largely in the price.

Take Tesla: it turned operating-profitable in 2020, and an entry right after the February 2021 filing returned +9.7% in a year — then −31.6% over three. The market had priced the trajectory long before the filing made it official. That pattern is the rule more often than the exception in the sample.

There is also a structural reason the hypothesis keeps failing at the median while looking alive at the mean. A profitability inflection is good news about the company, and good news about a widely followed company is the most efficiently priced commodity in markets. The events most likely to beat the median are the ones where the turn surprised — Palantir's 2023 profit landed in a year when consensus still modeled years of losses — while the most telegraphed turns (a SaaS company guiding toward leverage for four straight quarters) offer nothing to discover. In other words, the test measures whether a predictable quality milestone can serve as a surprise entry signal. Mostly, it cannot. The two tails that made the mean positive were, in fact, the two genuine surprises in the panel.

There is also a statistical lesson that justifies the pre-registration: the mean +3-year return for entries was +48.1% — comfortably positive — lifted entirely by fat tails (AppLovin +564.9%, Palantir +345.7%). "Buying the inflection beats buy-and-hold on average" is a true statement about this sample and a useless one, because averages lie in winner-take-most distributions. This is exactly why we registered medians and win rates as the primary metrics before the numbers existed.

What this result does not say

It does not say profitability is irrelevant. Profits arrived in all eighteen cases — real businesses got real. What failed is the entry signal, not the milestone. That distinction is the backbone of our standing position: the Blitz Index measures business quality, and this test says it should not be read as a price-timing tool.

It connects to what the ARK backtests already showed: the index is far better at separating winners from blowups than at naming entry points, and entry multiples dominate subsequent returns. A good company bought at the wrong moment is a familiar failure mode — this study just put a number on one version of it.

And the honest caveats, stated plainly because they were pre-committed. Nine observations is small, and the verdict is sensitive to definitions: excluding weakly fitting cases (one loss year before cyclicals turned, ABCL's temporary royalty spike) flips the +3-year median to +8.3% versus +4.8% for buy-and-hold. That sensitivity is logged as a registered follow-up — a cleaner event definition for a new test — not as a revision of this one.

The never-inflected group (CrowdStrike, Snowflake, Symbotic, IonQ, Rocket Lab and others) keeps the panel alive: their future inflections become future events. Fifteen of the eighteen events arrived in 2020 or later — the sample leans on the low-rate era — and the pre-2020 cases (Liberty Energy 2017, Solaris 2017, ServiceNow 2019) are the opposite tail. This is exploratory science, and we intend to treat it that way.

BlitzLab scores businesses against a scholar's standard — not investment advice. This article reports descriptive statistics of past patterns under a pre-registered methodology (H2, registered 2026-10-13, verdict: rejected). Past patterns are not predictions. No price targets, no buy/sell recommendations. Figures: GAAP, SEC-filed; estimates labeled as estimates.