A fall does not mean a recovery is due
Define the variable first: price level, return, margin, valuation multiple or company rank. The mean must refer to a comparable population and period. A rouble price without adjustments for share-count changes, distributions and inflation need not move around a constant economic level. The chosen baseline may have no reason to remain stable.
Two firms may have a stable 10% margin. A one-off order gives one 15% and a one-off shutdown gives the other 5%; when the shocks disappear, both return to 10%. Selecting only the former produces apparent deterioration, while selecting only the latter produces apparent improvement. This says nothing about a cheap stock because no prices were specified. A permanent licence loss is different: the former margin may not return.
Economic mechanisms need a time horizon
High industry profit may attract capacity and low profit may close capacity, but decisions take time and face entry barriers, debt obligations and cost differences. A plausible mechanism does not set a recovery date or guarantee every participant’s survival. Recovery of an industry after an issuer’s bankruptcy does not restore that shareholder’s investment.
State which metric should change, through which channel and what observation would falsify the hypothesis. A temporary shortage differs from an irreversible technology change. If the explanation changes its time horizon after every failed forecast, it becomes untestable. Show separate recovery and new-level scenarios in a valuation model.
What historical tests do and do not show
Poterba and Summers studied time-series components of returns and found dependence that differed by horizon, while also discussing limited test power. A later reappraisal by Kim, Nelson and Starz showed that the conclusion depended materially on subperiod selection; evidence for stable mean reversion was weaker than the initial interpretation. The disagreement is a warning about empirical sensitivity, not a defect in science.
These studies do not provide today’s entry price, waiting period or recovery promise. Read the return definition, countries, observation calendar and inference method. Overlapping multi-year intervals can look like many independent observations even though they reuse the same months. Table length is not automatically independent information.
The extrapolation error works both ways
It is dangerous to extend rapid growth forever and equally dangerous to promise recovery after a fall. Check whether institutions, competition, participant composition and measurement units remain comparable. If not, the old mean may be historical context rather than a forward reference.
A careful conclusion is conditional: if these causes persist, this direction is expected, but timing and magnitude are uncertain. Replace “when will my price return?” with “what are future cash flows worth using today’s information?” A personal purchase price does not create a recovery mechanism. Separate statistical regression from a required decline of every former leader.
A fictional obligations-and-demography model
A fully fictional pay-as-you-go system illustrates how a changed participant mix changes a financial flow; it is not a description of any pension programme. In state one, 100 workers contribute 10 each and 20 recipients receive 50 each: inflows and outflows are both 1,000. In state two, 80 workers contribute 10 and 25 recipients claim 50: inflows are 800, outflows 1,250 and the gap is 450 per period.
At a fixed benefit, balance would require 1,250 / 80 = 15.625 from each worker. At a fixed contribution, the possible benefit is 800 / 25 = 32. Neither alternative is advice to change real contributions or benefits. The issue is the changed numbers of payers and recipients, not a wrong arithmetic mean. A reserve could temporarily cover the gap, but its amount, return and exhaustion date are unspecified.
Do not equate this system obligation with a share. The shared lesson is conditional stability: an old P/E or other average remains informative only while its drivers remain comparable. Recalculate on the new base and identify missing data. Real programmes require current actuarial projections and applicable documents; issuers require their own obligations, capital and cash-flow scenarios.
Sources and limits
[1] Mean Reversion in Stock Prices: Evidence and Implications, NBER working paper 2343.
[2] Mean Reversion in Stock Prices? A Reappraisal of the Empirical Evidence, NBER working paper 2795. Historical studies and reappraisal; neither is a current price or trading instruction.