Wrecks and Rockets: the same reading, opposite ends

You probably remember MoviePass. Ten dollars a month, a film a day, and a long public argument about whether the arithmetic could possibly work.

Weeks before that offer existed, at the end of June 2017, this study had already taken its reading on the parent company. Helios and Matheson carried two of them: a share price moving more wildly than almost anything else listed in the United States, and a balance sheet in the weakest third of its sector. Two years later the holding was worth nothing.

Which looks like a clean win for the obvious interpretation. A jumpy price means trouble.

Except that in the same month, in the same box, on the same two readings, sat a diagnostics company called CareDx. Over the following two years it gained thirty-one times its value.

Everybody knows what a jumpy share price means. Something is wrong. The market has smelled trouble and is repricing it in public.

A lot of that is right, and it is worth starting there.

We sorted 1,279,541 company-months of US listings by how much each share price had been moving, then waited two years and looked at what happened. Holding the balance sheet constant at the middle of the pack, the typical outcome slides from +16.2% in the calmest group to −10.4% in the wildest. The pattern is real. On average, jumpier means worse.

If that is all you wanted to know, the claim holds. The rest of this is about how much that average is hiding.

The average is hiding almost everything

Here is what the average does not tell you.

In the calmest group, the distance between a bad outcome and a good one was about 106 percentage points. In the wildest group, that distance was 250 percentage points.

Both ends stretched. The bad outcomes got much worse, and the good outcomes got much better. Wild companies were not all sliding toward the same place. They were scattering.

That scattering is not a footnote. The single group with the highest rate of five-fold gains in the entire study is the wildest, weakest corner, at 2.89%. It is also the group with the highest rate of near-total loss, at 19.06%. The wrecks and the rockets sit in the same bucket, and price movement alone will not separate them.

A jumpy share price tells you the range of things that might happen is wide. It does not tell you which one.

So what does tell you which way?

If price movement gives you the width, something else has to give you the direction. We tried the obvious candidate: the company’s own financial condition.

For every reading we also scored how solid the books looked at the time, using what had actually been filed by that date. No hindsight, no restated figures. Then we split each turbulence group into thirds: strongest finances, middle, weakest.

Take one of these companies at a random moment. Here is how the next two years went.

The difference between a wild company with solid books and a wild company with weak books is 24.1 percentage points in the typical outcome. In the calmest group the same gap is 6.3 points. The balance sheet matters more the wilder the price gets, and all of it comes from information that was public at the time.

The part we did not expect

Now look at the two extremes rather than the middle.

The chance of losing almost everything goes from about 1 in 19 to about 1 in 5 as the books weaken. That much you would guess.

The surprise is the other bar. The chance of a five-fold return is higher in the weak group, not lower. 2.89% against 1.55%.

So weak books did not simply remove the good outcomes. They changed the mix. Going from the strongest books to the weakest, the chance of a five-fold gain roughly doubles. The chance of near-total loss rises three and a half times.

The clearest way to see it is to put the two ends against each other. Among the wild companies with the strongest books, there were about 3.4 near-total losses for every five-fold gain. Among those with the weakest books, 6.6. Both ends get more crowded as the books weaken. The bad end fills up about twice as fast.

The same box, two years later

Names make this concrete. Every company below was measured in mid-2017 and sat in the same box: Extreme turbulence, weakest third of balance sheets. The outcome column is what had actually happened by mid-2019.

The same box, sorted the other way.

CareDx and Helios and Matheson carried the same two readings in the same month. Over the next two years one gained thirty-one times its value and the other went to zero. Helios and Matheson is the company that owned MoviePass.

These are the two ends of that box, selected as ends. They are not typical members. The typical member of it lost 32.3%.

Where the balance sheet shows up is the floor. Among the wild companies with the strongest books the worst outcomes stop around ninety to ninety-nine percent, names like Pier 1 Imports and InVivo Therapeutics. Among the weakest, the bottom of the list is a column of total losses. Weak books did not remove the ceiling. They lowered the floor.

Sorted by price movement alone, the tenth-to-ninetieth percentile range runs 76 points in the calmest band and 212 points in the wildest. Split the wildest band by the books and the weakest third is the widest of all at 240 points, with both a lower floor and a higher ceiling than the strongest third. That is the same result as the rates above, seen as a distribution rather than as two percentages.

The names and the chart in this section come from a single 2017 measurement cohort of 2,763 companies, rebuilt so that every row can be checked by name. The rest of this study uses the full 1,279,541 company-months. The cohort reproduces the study’s ordering; it is a much smaller sample and its figures are not the study’s figures.

The obvious objections

Cut data into boxes and you can usually find something. Four turbulence groups here, three balance-sheet groups there, and the boundaries are ours. A result that only shows up at one particular set of cut points is not a result, it is a coincidence with good presentation.

So we did it again without the boxes. Scoring both measures on a continuous scale and sorting every reading into ten groups produces a clean slide from +18.0% at the calm and solid end to −25.7% at the wild and weak end, with no reversals anywhere in between. That is a 43.7-point spread, against 24.8 for the four boxes.

The pattern is in the data, not in the way we cut it. It is also stronger without the boxes than with them, which is the opposite of what a data-mined result looks like. It is not flattering to us, either, and we come back to that below.

It could be a small-company story. Tiny companies are jumpy, fragile and occasionally spectacular, so the whole thing could be a microcap effect wearing a costume. We re-ran the identical measurement on the 3,000 largest companies by market value in each month, which removes the microcaps entirely.

The pattern survives and it gets weaker. Among the 3,000 largest, the slide runs from +18.3% to −4.8%, a spread of 23.1 points against 43.7 on the full population. The chance of losing half your money still climbs from 3.8% to 28.8%.

We are reporting that plainly rather than burying it. Roughly half the effect on the full population comes from companies too small for most people to own. The other half does not.

It could be a sector story. Biotechs are naturally jumpy and utilities are naturally calm, so a measure of financial condition could just be a measure of industry. Every company is scored against other companies in its own sector in the same month, never against the market as a whole. A biotech is compared with biotechs. That comparison is built into the measure rather than argued for afterwards.

The part that failed is ours

Filter Lab sorts companies into four named homes. Steady, Quiet, Wide and High-risk. Those four boxes span 24.8 points of two-year outcome, best to worst. Rank the same companies on the same two measures and cut them into four equal groups instead, and those span 27.4. The labels keep 91% of the separation.

An earlier version of this section said they threw away 43%. That compared our four named homes against ten equal groups, and more groups always reach further into the tails. The number was measuring our own arithmetic rather than the market, which is precisely the failure this study is about, sitting in the one section that admits a fault. It was caught by a reader asking what it meant.

Ten groups do show where the resolution fails, and what they show is narrower and more useful. Rank every reading and cut it into ten, and the typical two-year outcome runs 18.0, 16.7, 16.4, 14.1, 10.8, 7.8, 4.9, 1.1, then −6.3, and then −25.7. A 19.4-point fall in a single step, 44% of the entire range at one end of it. Our worst home holds 346,194 rows, a quarter of the universe. You cannot point at the worst tenth with a bucket that holds a quarter.

So the boxes are the right shape and the wrong resolution, at one end and only one end. We are keeping the plain-English labels, because that is how anybody actually reads a market. What they cannot do is isolate the tail where nearly half the separation lives.

So the honest summary is narrower than we would like, and more useful than it sounds:

Price movement tells you how wide the range of outcomes is. The balance sheet tells you how lopsided that range is. Neither one tells you which end you are going to get.

What this does not say

It is not a prediction, and it is not advice. Everything here describes what already happened to a large group of companies. Nothing here says what will happen to any particular company, and nothing here suggests buying or selling anything.

Averages are not destinies. A typical outcome of −31% means half did worse and half did better. Plenty of companies in the worst corner recovered completely.

Companies are named here only in closed windows. Every name in this piece sits in a reading taken in 2017 with an outcome that finished in 2019, so naming it reports what happened rather than judging what will. No company is named from a live reading, and nothing here says anything about any of these businesses today.

This was exploratory. We did not specify these particular cuts in advance. The continuous check is there because a result that only appears when you draw the boxes a certain way is not a result.

The last two years are measured over a shorter window. A reading gets whatever forward price exists up to 24 months out, so readings near the end of the panel are followed for less than the full two years. That compresses the most recent rows toward smaller numbers in both directions, and it should be re-run with a hard 24-month cutoff.

Financial companies and property companies behave differently. Ranking against sector peers helps, but a current ratio means something different for a bank than for a manufacturer. Banks and real estate firms are a known weak spot, not a solved problem.

One company appears many times. The unit here is a company-month, not a company. A firm that spent three years in the wildest band contributes 36 rows, so these are rates across readings rather than counts of businesses.

Method

US-listed common stock, 1,279,541 company-months. Price movement is the rolling 90-day variance of daily returns, annualised, then averaged over the prior trading year and cut into four fixed bands. Financial condition is a blend of return on assets, current ratio, debt against assets inverted, and free cash flow against assets, taken from filings as they stood on the date in question and ranked against sector peers in the same month. The combined ranking used for the ten-group check scores both measures against sector peers; the four bands are fixed thresholds rather than sector-relative, which is why the two are reported separately.

Forward returns run up to 24 months. Bankruptcies are carried at a 100% loss rather than excluded. Acquisitions are carried at the last traded price and treated as survivals. Delisted companies are kept in. Companies trading below one dollar at the measurement date are excluded. Medians throughout, so one extreme name cannot carry a group.

Filter Lab measures market structure and does not make recommendations. Turbulence measures the range of a company’s movement, not its direction. Descriptive only: not advice, not a recommendation, not a forecast.