CompanyGraph
An extreme result can be followed by a less extreme one because the first observation included noise that is unlikely to repeat—not because the system is being pulled back.
What “regression” means here
Suppose a company’s sustainable margin is 12 percent, but one year combines that operating ability with an unusual input-price benefit and a favourable product mix. The reported margin is 20 percent. If the next year removes the temporary benefit, the result may be closer to 12 percent even without managerial failure. The same logic applies to an unusually poor result: a one-off disruption can disappear without a permanent improvement in the underlying business.
NIST’s statistical glossary treats regression as a method for relating an outcome to explanatory variables. Regression to the mean is the conditional expectation that a selected extreme will be less extreme on a later measurement when the first observation contains random or measurement variation. It is not the claim that every series returns to its average.
The mathematics does not assign a cause
Imagine a manager whose true ability produces an average result, plus a random shock. If investors select the manager after an unusually high result, the next result is expected to be lower because the same favourable shock is unlikely to recur. The lower result does not prove that the manager deteriorated. If the selection was based on a poor result, the follow-up may improve for the same statistical reason.
The effect is stronger when measurement noise is large relative to persistent signal, when the sample is small, and when selection is made at an extreme. It is weaker when the process is stable and measured precisely. A performance metric can also move because the underlying mean changes; regression analysis cannot decide whether a new technology, market, or regulation has reset the baseline.
Investment performance is a bounded example
De Bondt and Thaler’s long-term return-reversal study found a historical tendency for prior extreme losers to outperform prior extreme winners over the horizon and sample they examined. That result is evidence about a defined portfolio construction and period. It is not a general rule that every losing stock will recover or that a winner must fall.
Fund performance, quarterly earnings surprises, sports results, and customer growth can show similar selection effects. But a subsequent decline can also reflect fees, style change, capacity, competition, or bad decisions. The statistical concept tells the investor to ask how much of the initial result was likely to repeat; it does not settle the explanation.
The deep-value configuration is observable: companies priced well below book value while current assets exceed current liabilities by a wide margin and the equity ratio sits high for the industry.
Inverted P/B With Liquidity And Equity Ratio
Inverted P/B is high (price below the P/B scale) while current assets exceed current liabilities by a wide margin and equity is in the upper part of its industry's equity-to-assets range
A constructed portfolio's historical result is not a forecast for any member. Membership records a price against a book value today, nothing more.
Business metrics can have different baselines
- Margins. A temporary commodity price or mix benefit can fade, while a new process or patent can raise the sustainable margin.
- Growth. A small base, a launch, or an acquisition can produce an extreme growth rate that naturally moderates as the denominator expands.
- Returns. A high return on capital can be partly cyclical or leverage-assisted; compare the operating return and the financing structure.
- Quality events. A single recall or outage can make a weak year look worse than the long-run process, but repeated failures can indicate that the baseline has deteriorated.
- Valuation. A multiple can fall toward a historical range, or the range can reset when rates, duration, or business quality changes.
The analyst must distinguish regression from genuine improvement or decline. If a poor result is followed by a better one, ask whether the original shock ended, management changed the process, demand recovered, or the measurement merely returned toward its prior distribution.
Do not confuse regression with causation
A school, hospital, or sales team selected for an unusually poor result may improve on the next measurement even if no intervention works. Conversely, an intervention can be effective while the next result still worsens because a new shock arrives. Before-and-after comparisons that select on an extreme are vulnerable to this regression effect.
In investing, a new CEO who arrives after a trough may receive credit for statistical recovery. A manager who inherits a peak may be blamed for normalization. The attribution requires a comparison group, a mechanism, and evidence that the action changed the process rather than coincided with a natural reversal.
What investors can test
- Define the measured outcome, reference population, selection rule, and time horizon.
- Separate persistent factors from one-off events, measurement error, mix, and leverage.
- Compare the selected company with a relevant peer or control group rather than with an unexamined average.
- Test whether the business process, customers, competitors, or regulation changed the underlying baseline.
- Do not pay for an extreme result as if every favourable condition will repeat, but do not assume a weak result must recover.
- Use subsequent evidence to attribute improvement or decline; the direction alone is not proof of management skill or failure.
Regression to the mean is a discipline against over-attributing an exceptional observation. It keeps the persistent capability, temporary conditions, and uncertainty separate so that the next result can be interpreted rather than celebrated or blamed automatically.