|
The random walk hypothesis is not so much a hypothesis as
it is a model that has been found to be surprisingly useful for
describing the behavior of prices in various markets, including
major equity, fixed income and commodities markets. It states that
price series do not exhibit predictive patterns over time but can best be
described with a random walk. Accordingly, the random walk hypothesis is
a rejection of technical analysis.
An
early version of the random walk hypothesis was proposed by
Louis Bachelier (1900) in his famous doctoral thesis Théorie
de la Spéculation. Bachelier
studied the market for forwards and options on French government
bonds, discovering a number of
important results. As part of that work, he discovered the mathematics
of Brownian motion—five years before Albert Einstein independently did
so. With regard to the markets, Bachelier noted
[1]
The influences that determine the movements of the exchange are
innumerable; past, current and even anticipated events that often have
no obvious connection with its changes ... it is thus impossible to hope
for mathematical predictability.
He went on to conclude that, if the market's movements cannot be
predicted,
The mathematical expectation of the speculator is zero.
Bachelier's thesis was
decades ahead of its time. It was ignored for over a half century, but
other researchers, acting independently, started to draw similar
conclusions.
Holbrook Working was an agricultural economist with
Stanford University’s Food Research Institute. In a 1934 study, he compared
time
series of historical commodity price changes to time series of random
numbers. He wanted to determine what non-random price patterns might be
exploited by traders to realize speculative profits. Using statistical techniques, he was unable to distinguish
the series of price changes from the series of random numbers. He
concluded that there were no predictive patterns in the price
changes—that the prices were entirely random. He showed graphs of the
time series to professional commodity traders. They too were unable to
distinguish the series of price changes from the series of random
numbers.
Maurice
Kendall was one of the great statisticians of the 20th century. In
1953,
he published a ground-breaking empirical study of weekly changes in
nineteen indices of British industrial
share prices and in
spot prices for New York cotton
and Chicago wheat. His goal was to advance the field of technical analysis
by introducing statistical rigor. He was startled to find that the
random component of prices swamped any
autocorrelations. Frustrated, he
concluded [2]
The series looks like a wandering one, almost as if once a week the
Demon of Chance drew a random number from a symmetrical population of
fixed dispersion and added it to the current price to determine the next
week's price.
Working's and Kendall's conclusion that past price data could not
be used to predict future prices effectively rejecting the practice of
technical analysis. But traders can base decisions on information other
than past price data. An equities trader might, for example, base trades
on a firm's past earnings growth, its price/earnings ratio, how
experienced its management team is, prospects for the firm's industry,
and more. This is the realm of fundamental analysis. The studies by
Working and Kendall did not address fundamental analysis, but earlier
work by Alfred Cowles had.
Alfred Cowles 3rd was the son of a wealthy
businessman. He subscribed to a number of investment newsletters during
the 1920’s and 1930’s to help manage his family's fortune. Following the
crash of 1929, he looked back and realized that none of the newsletters
had reasonably predicted the crash or subsequent events. Wondering if
the poor performance was coincidence or typical, Cowles enlisted the aid
of economists to conduct a massive study. Using an IBM punch-card
machine—a precursor of the computer—they analyzed the historical
investment decisions of 25 investment newsletters, 16 stock tip services
and 12 fire insurance companies.
Cowles
published the results in a 1933 paper entitled "Can Stock Market
Forecasters Forecast?" He summarized his conclusions with a 3-word
abstract: "It is doubtful." Most of the practitioners had underperformed
the broad market. Their combined results had underperformed as well. The
few practitioners who had done well did so modestly, so there was no
indication that their performance was due to skill instead of luck. Cowles also
experimented with series of forecasts constructed by randomly drawing
cards from a deck. He found that the best random series of forecasts
outperformed the best practitioners—and the worst series of random
forecasts outperformed the worst practitioners.
Cowles did much to further research into the performance
of stock pickers. In 1932, he launched the Cowles Commission, which has
sponsored considerable research. In 1933, he helped found the prestigious journal
Econometrica. He launched an equity index that became today's S&P
500. In 1944, he published a large follow-up study to his 1933 study of
stock pickers' performance. It drew a similarly dreary conclusion.
Cowles' efforts went unnoticed by brokers and
investment managers, who made their living off investors believing it was possible to outperform the market. But academics
took
note. By the 1960s, there was an active literature emerging on what was
then called the random walk hypothesis. In 1965, Paul Cootner published an
influential book called The Random Character of Stock Market Prices.
This pulled together reprints of important articles up to 1963,
including Cootner's own work and the early works of Bachelier and Kendall.
Bachelier's original version of the random walk
hypothesis was crude by today's standards. It had prices follow an
arithmetic random walk with zero drift. The modern version developed out
of the work of multiple researchers, including those already mentioned,
as well as Osborne (1959), Moore (1960), Alexander (1961) and Granger
and Morgenstern (1963). It states that the log returns follow an arithmetic
random walk with a drift reflecting the long-term return from equity
investment. Stated another way, prices follow a
geometric random walk
with drift.
The random walk hypothesis is more an empirical
observation than a theoretical result. Fundamentally, it is an empirical
observation that price series are well modeled with a random walk.
However, researchers did offer a theoretical explanation for why prices
should follow a random walk. They noted that prices change in response
to news items—earnings reports, releases of economic indicators, merger
announcements, etc. If news items are assumed to arise independently
(the relative probabilities of upcoming news being good or bad is
unaffected by whether recent news has been good or bad) then price
changes should be independent. Also, the volume of news items affecting
a price is sufficiently large that the
central limit theorem applies,
and price changes over any discernible period should be approximately
normal. Somewhat after
the fact, Samuelson (1965) and Mandelbrot (1966) rigorously formalized
this theoretical justification of the random walk hypothesis.
In the early 1960s, the literature of the random walk
hypothesis took two new directions. The first would extend results
as an efficient market hypothesis. The other looked for flaws in the
random walk model. Two obvious flaws, which were noted early on,
were the fact that log price changes were
leptokurtic,
rendering the normal distribution an imperfect representation, and
heteroskedastic.
Neither flaw affords technicians trading opportunities,
but researchers sifted through massive volumes of historical price data,
looking for flaws that might. Their published results formed the
literature on market anomalies which ultimately contributed to the
dubious field of behavioral finance.
|
|
 |
|
|
|
|
 |
 |
|
|
|
 |
|
Alexander, Sidney (1961). Price movements in speculative markets:
trends or random walks, Industrial Management Review, 2
(2), 7-26. [Reprinted in Cootner.]
Bachelier,
Louis (1900). Théorie de la Spéculation, Annales Scientifique de
l'École Normale Supérieure,
3e série, tome 17, 21-86.
[English translation in Cootner;
original French with a more recent English translation in
Davis and Etheridge.]
Cootner,
Paul (1965). The Random Character of Stock Market Prices ,
Cambridge: MIT Press.
Cowles,
Alfred 3rd (1933). Can stock market forecasters forecast?,
Econometrica, 1 (3), 309-324.
Cowles,
Alfred 3rd (1944). Stock Market Forecasting,
Econometrica, 12 (3 & 4), 206-214.
Davis, Mark and Alison Etheridge (2006).
Louis Bachelier's Theory of Speculation , Princeton: Princeton University Press.
Granger,
Clive and Oscar Morgenstern (1963) Spectral analysis of New York
stock market prices, Kyklos, 16 (1), 1-27. [Reprinted in Cootner.]
Kendall,
Maurice (1953). The analysis of time series, Part 1: Prices,
Journal of the Royal Statistical Society, Series A
(General), 116 (1), 11-34. [Reprinted in Cootner.]
Mandelbrot,
Benoit (1966). Forecasts of future prices, unbiased
markets, and martingale models, Journal of Business, 39
(Special Supplement, January), 242-255.
Moore, Arnold (1960). Some characteristics of changes in common
stock prices, doctoral thesis, University of Chicago. [Reprinted in Cootner.]
Osborne,
M. F. M. (1959). Brownian motion in the stock market,
Operations Research, 7, 145-173. [Reprinted in Cootner.]
Samuelson,
Paul A. (1965). Proof that properly anticipated prices fluctuate
randomly. Industrial Management Review, 6 (2), 41-49.
Working,
Holbrook (1934). A random-difference series for use in the
analysis of time series, Journal of the American Statistical
Association, 29 (185), 11-24. |
|
|
|
 |
 |
Ads by Contingency Analysis
|
|
|
 |
|
|
|
|