1984issue C051-15
When random walks fail to explain market averages
Treat the random-walk claim as a testable baseline. Independent drawings can mimic single-name charts, but summing those drawings need not recover the cyclical swings of a market average.
- A random-walk hypothesis test checks whether later price changes can be treated as independent of earlier ones, so a buy-and-hold or shuffled path is the right baseline.
- Monte Carlo simulation can produce wandering monthly paths and familiar formations, and unlabeled real one-year charts were not picked as often as chance would suggest.
- Summing many independent series, including groups of 30 and 500 paths, tended toward a flatter line and did not reproduce the industrial average's dominant cyclical swings.
- A runs test on 120 monthly Dow changes found 47 runs against about 60 expected under independence, a deficit treated as evidence against a pure random walk at that interval.
Independence as a baseline
The random-walk case is framed as independence of successive price changes. Prior moves are said to give no information about the next change.
A supporting rationale is that newly available information is used almost at once to revise prices, leaving later changes unexpected and therefore random-looking. That is the efficient-market rationale: leftover changes look unexpected, and therefore random.
A random-walk hypothesis test is a check of whether later price changes can be treated as independent of earlier ones, so a simple buy-and-hold or shuffled path is the right baseline.
Single names and random drawings
Monte Carlo simulation builds synthetic price series from a stated random process so those series can be compared with observed charts or indexes.
In one cited comparison, analysts reviewing unlabeled one-year charts of five real stocks versus random simulations did not pick the real series as often as chance would suggest.
Card-dealt random monthly paths over a ten-year window produced wandering curves and familiar-looking formations, including a rising head-and-shoulders shape on one series. On the same long span, some listed-stock charts resembled those random drawings while others did not. The resemblance looked stronger on shorter spans with weekly or daily increments.
What happens after index aggregation
Index aggregation adds many individual series into an average or index so shared swings can appear even if each series was generated independently.
A computer Monte Carlo using increments bounded from +6 to -6, then summed across 30 series, did not reproduce the dominant cyclical swings seen in the industrial average plotted on the same scale.
Summing 500 such random series and scaling the total produced a much flatter path. That result was used to support the claim that wider random aggregation tends toward a nearly horizontal line.
A runs count on monthly changes
A runs test counts consecutive same-sign moves to judge whether a series switches direction as often as an independent process would.
A runs count on 120 monthly Dow changes from January 1974 through December 1983 found 47 runs against about 60 expected under independence. That deficit was treated as evidence against a pure random walk at that sampling interval.
How to read the comparison
Editorial reading: a random-looking single name does not settle the question for an average. Compare the generated independent paths with both the constituents and the summed index before treating independence as the working description of the market average.
All readings on this track · 10 readings
- 1984When random walks fail to explain market averages
- 1988Building probability-density adaptive weights and a chi-square trend filter
- 1990Withhold a random-walk label until grouping and lookback are checked
- 1991Constructing a variable-lookback random walk trend filter
- 1992Ranking futures trend tendency against a random walk
- 1995A normalized dispersion screen for trend-following
- 1997Why failed random-walk tests still need careful Monte Carlo work
- 2007Constructing a random-walk baseline from ordered returns
- 2011Constructing a cycle-free trend filter against an ATR-scaled random walk
- 2013How to evaluate a hybrid growth and money-flow scan