See discussions, stats, and author profiles for this publication at: [Link]

net/publication/5147308

Dollar Cost Averaging

Article in European Finance Review · December 2005

DOI: 10.1007/s10679-005-4999-x · Source: RePEc

CITATIONS READS
49 4,591

3 authors, including:

Michael J. Brennan
University of California, Los Angeles
167 PUBLICATIONS 23,826 CITATIONS

SEE PROFILE

All content following this page was uploaded by Michael J. Brennan on 05 March 2016.

The user has requested enhancement of the downloaded file.

 Anderson Graduate School of Management
– Finance
UC Los Angeles

Title:
Dollar Cost Averaging

Author:
Brennan, Michael J, UCLA
Li, Feifei, UCLA Anderson School
Torous, Walt, UCLA

Publication Date:
06-24-2005

Publication Info:
Finance, Anderson Graduate School of Management, UC Los Angeles

Permalink:
[Link]

Abstract:
Dollar Cost Averaging is a strategy for purchasing equity securities that is widely recommended
by professional investment advisors and commentators, but which has been virtually ignored
by academic theorists and textbook writers. In this paper we explore whether the strategy is
but another instance of irrational behavior by individual investors, or whether it is an investment
heuristic that has survival value in an environment in which security prices exhibit mean reversion
behavior that has only belatedly been recognized by academic theorists. Our evidence supports
the view that the individual investors who follow this strategy in purchasing individual stocks to
add to an existing portfolio are better off than if they followed the 'rational' strategies traditionally
recommended by academics.

eScholarship provides open access, scholarly publishing

services to the University of California and delivers a dynamic
research platform to scholars worldwide.
 Dollar Cost Averaging

Michael J. Brennan

Feifei Li
∗
Walter N. Torous

June 24, 2005

∗ Michael Brennan is Emeritus Professor, Feifei Li is a doctoral candidate, and Walter Torous is Professor Finance

at the Anderson School, UCLA. Corresponding Address: Brennan: The Anderson Graduate School of Management;
University of California, Los Angeles; 110 Westwood Plaza; Los Angeles, CA 90095-1481. We thank the editor and the
referees for helpful comments on an earlier draft. The paper was presented at a conference organized by La Fundacin
Ramn Areces in Madrid, 2005.
 Abstract

Dollar Cost Averaging is a strategy for purchasing equity securities that is widely recom-

mended by professional investment advisors and commentators, but which has been virtually

ignored by academic theorists and textbook writers. In this paper we explore whether the strat-

egy is but another instance of irrational behavior by individual investors, or whether it is an

investment heuristic that has survival value in an environment in which security prices exhibit

mean reversion behavior that has only belatedly been recognized by academic theorists. Our

evidence supports the view that the individual investors who follow this strategy in purchasing

individual stocks to add to an existing portfolio are better off than if they followed the ‘rational’

strategies traditionally recommended by academics.

1 Introduction

Practical or tacit knowledge typically precedes scientific knowledge. Crops were rotated long before

the chemical basis of the practice was understood. Men learned to fly before aeronautics was well

understood. Extracts of willow were described by Hippocrates as a pain remedy well before Bayer

first synthesized aspirin. Acupuncture was in use for millennia before Nixon’s 1972 trip to China

made the practice scientifically acceptable in the West. And these are only a few examples of the

practical wisdom that is embodied in pre-scientific prescriptions. Therefore, given the relatively brief

period of scientific study of financial markets, and the controversy that surrounds the interpretation

of many of the findings, it is not surprising to find that a good deal of financial practice is still

governed by pre-scientific heuristics or maxims. Examples of such maxims include, ‘buy on the

rumour, sell on the news’, ‘cut your losses and let your profits run’, and the ‘60-40’ and ‘100 minus

your age’ rules for asset allocation. Just as we would expect the practices of folk medicine that

have survived to possess some curative properties, or at worst to be harmless, so we might expect

these investment maxims also to have some value or, at worst, to be harmless, unless they conflict

with strongly supported scientific evidence. In this paper we are concerned with the ‘dollar cost

averaging’ (DCA) heuristic.

Financial markets are somewhat different from agricultural crops, flying machines and human

bodies as objects of scientific study, since the properties of markets are affected by the beliefs

and practices of investors in the markets, and these beliefs and practices are influenced in turn

by scientific studies of the markets.1 This feedback between human decisions, the properties of

markets, and the resulting profitability of the decisions has led some scholars to study decision rules

for participants in financial markets that have survival value against other decision rules.2 This is a

very different approach from the standard rational expectations approach in which all investors3 are

assumed to have full information about the structure of the economy and about each other’s beliefs.

1 Brennan (2004) argues for example that the efficient markets hypothesis and popular knowledge of the historic

equity premium led investors to pay excessive prices for stocks in the 1990’s on the basis that stocks beat bonds in
the ‘long run’.
2 Blume and Easley (1992), Routledge (2001), Lo(2002), Hens et al. (2003).
3 With the possible exception of some ‘noise traders’

1
Behavioral finance is yet a third approach that has developed of late in reaction to the inability of

standard rational expectations models to explain a number of stylized facts about financial markets.4

Proponents of this paradigm believe, not only that a significant proportion of investors are irrational,

but also that their irrationality has important implications for market prices; however, the paradigm

has not been exploited to derive formal prescriptions for investors operating in these markets.5

In contrast to the behavioral position, this paper takes seriously the possibility that the practical

maxims or heuristics that we have inherited are not manifestations of investor irrationality, but are

actually helpful for investors who operate in markets that do not conform to the simple descriptions

of standard asset pricing models, even if the reasons for the failure of these models remain obscure.

We examine the value of the dollar cost averaging heuristic by randomly choosing investments dates

and securities from the CRSP database over the period 1926-2003 and comparing the certainty

equivalent wealth levels yielded by lump sum and dollar cost averaging investment strategies for an

investor with iso-elastic utility.

We find, first, that, for an investor who is purchasing a diversified investment portfolio of common

stocks represented by the CRSP value-weighted or equal-weighted indices, the DCA strategy, carried

out over implementation periods of from one to six years, outperforms the lump sum investment

strategy for all except the most risk tolerant investors.6 This seems to be due to the lower level of

risk associated with the DCA strategy since the expected returns are, not surprisingly, higher for

the lump sum strategy. However, we also find that both strategies are outperformed by a simple

strategy of investing 50% of wealth in stocks and 50% in cash, and rebalancing monthly to maintain

the proportions.7 As it turns out, the 50:50 strategy has almost the same expected return as the

DCA strategy, but lower risk. Therefore we find only mixed support for a DCA strategy when

applied to the purchase of a diversified portfolio of common stocks.

4 See also the theory of ‘rational beliefs’ of Kurz (1994,a,b).

5 However, a number of investment funds claim to be operated on behavioral principles. For example Fuller
and Thaler Asset Management utilizes a ‘bottom-up investment approach that combines fundamental research with
insights from behavioral finance...’. ([Link]
6 Of course, both strategies are outperformed by the ‘optimal’ investment strategy which bases the allocation to

equities on the realized sample moments of the equity returns.

7 Brennan and Torous (1999) show that rebalancing strategies are slightly more efficient than buy and hold

strategies.

2
 Secondly, when the DCA strategies are compared with the lump sum investment strategy for the

purchase of a single stock which is the investor’s sole investment, the DCA strategy overwhelmingly

dominates the lump sum strategy. The reason for this is the lower risk of the DCA strategy, and it

is worth noting that for all except the most risk tolerant investors, the certainty equivalent wealth

per dollar of initial investment is less than one - investment in a single stock portfolio is, in general,

not an economic proposition - even when it is done by DCA.

Thirdly, given the inefficiency of a single stock portfolio, it is natural to consider the merits of

DCA when applied to the purchase of an additional stock by an investor who already holds a portfolio

of common stocks. When we analyze the case in which the investor is assumed to hold the (value- or

equal-weighted) market portfolio and to purchase a marginal amount of a randomly chosen additional

stock, the DCA strategies dominate the lump-sum purchase strategy in all cases for investment

horizons up to 48 months. If one were to choose an investment strategy for ‘everyman’ assuming

that he holds the (value- or equal-weighted) market portfolio, then a DCA strategy executed over

36-48 months would be the preferred strategy. This result is re-inforced when the investor is assumed

to hold initially a diversified portfolio of common stocks that is optimally levered given his level of

risk aversion. Now DCA dominates the lump sum investment strategy for an investor who holds the

value weighted market portfolio for all levels of risk aversion and implementation periods up to 60

months, and the advantage of DCA is greatest for the 36 month DCA strategy where it dominates

the lump-sum strategy by a margin of 3-6% depending on the level of risk aversion and the random

method used to select securities. When the initial portfolio is the optimally levered equal weighted

market portfolio, the advantage of DCA disappears or is even reversed.

While, it is hard to determine the precise features of the joint stochastic process of common

stock returns that accounts for this success of DCA, simulations reveal that transient departures

of common stock prices from their fundamental values in the manner suggested by Poterba and

Summers (1988) can give rise to an advantage to DCA which is similar to that we find in the data.

In summary, we find that dollar cost averaging, when applied to the addition of a stock to an

existing portfolio, has benefits that have not received attention in the academic literature despite the

3
fact that investment practitioners have tended to advocate the strategy for many years. It appears

to be a case in which practical wisdom has discovered what theoretical knowledge has yet to attain.

2 Evidence on the Effectiveness of Popular Investment Strate-

gies

It is important to note that popular prescriptions for investment success are standardized prescrip-

tions for ‘everyman’, that generally do not take account of the details of the individual situation

and, in particular, do not take account of the tastes of the individual as represented, for example,

by the risk aversion of his von Neumann-Morgenstern utility function. In contrast, modern scientific

investment rules8 are predicated upon the precise specification of investor utility functions, even

though it may be very difficult in practice to elicit such utility functions from the individual. Thus

we should not expect the popular prescriptions to be ‘optimal’ for any given utility function, though

we might expect them to be robust in the sense that they are reasonably close to optimal for a broad

range of utility functions if the utility functions are ‘reasonable’.

Perhaps the most fundamental decision faced by the investor is the appropriate allocation to risky

securities. Early modern financial theory9 makes this decision a complex function of the distribution

of returns and investor tastes, while the simple 60-40 rule prescribes a simple 60% allocation to

equities and 40% to cash. Brennan and Torous (1999) show that this simple prescription attains in

practice a large fraction of the potential gains from investing in equities for a broad range of risk

aversion. More generally, DeMiguel et al. (2005) show that when there are N asset classes, a simple

1/N asset allocation rule generally outperforms more sophisticated rules bases on financial theory.

A more sophisticated variant of the 60-40 rule is the prescription that the percentage of his

wealth that an investor should allocate to equities is equal to 100 minus his age in years.10 This age

dependent allocation rule conflicts with early analytic models of dynamic investment planning such

as that developed by Samuelson (1969) which prescribe that, at least for standard iso-elastic utility

8 See
for example Campbell and Viceira (2002)
9 Standardmean-variance analyses of the Markowitz (1959)-Sharpe (1963) type.
10 This rule is advocated for example by Malkiel (1996, p 418).

4
functions, the optimal equity allocation is age independent.11 However, more sophisticated analytic

models such as Bodie et al. (1992) and Cocco et al. (2004), which take account of the depreciating

endowment of human capital, tend to validate the principles, if not the parameters, of the heuristic

age dependent rule. For example, Cocco et al. (ibid.) show that for their base case parameters, the

100 minus age rule cuts the welfare loss of the simple Samuelson (ibid.) rule by around 58%.

In a paper that stimulated broad interest, Canner et al. (1997) questioned the more detailed

asset allocation rules that have been proposed by popular investment advisors. They noted that, in

contrast to implications of the standard Tobin Separation Theorem, the investment advisors recom-

mended that more risk averse investors have proportionately larger allocations to bonds. However,

the Tobin Separation Theorem is derived in a static setting and, as Brennan and Xia (2000) among

others have shown, the advice of the popular investment advisors is easily rationalized in a dynamic

setting when the interest rate is stochastic. In this case, the popular advice seems to have been based

on a more sophisticated, albeit implicit, model than the model its academic critics were employing.

The popular advice to investors that they should cut their losses and let their profits run finds

no support in the classical scientific theory of investment unless it is based on tax considerations.12

Indeed, the random walk version of the efficient markets hypothesis and standard mean variance

analysis would rather suggest the opposite: a rebalancing strategy in which winners are partially

liquidated and losers further purchased in order to maintain portfolio efficiency. However, the

evidence of price momentum that began to appear in the early 1990’s, starting with Jegadeesh and

Titman (1993), raised the possibility that perhaps the popular maxim was not without merit, even

in the absence of tax considerations. More recent work by Odean (1998) has established that, not

only is it more profitable to follow the strategy implied by the maxim, but that individual investors

have a tendency to do the opposite - that is, to sell winners and to hold on to losers. Thus, once

again, the folk-lore strategy, although contrary to the modern ‘scientific’ view, has been vindicated

by more careful analysis of the actual behavior of prices in securities markets.

11 Samuelson (1989) has been at pains to underline the error of an equity allocation that reduces with age which he

attributed to a misunderstanding of the nature of risk.

12 See Constantinides (1983) and Dammon et al. (2001).

5
 In summary, there is evidence that popular heuristics or maxims for personal portfolio manage-

ment have in some cases anticipated the findings of scientific studies of financial markets. One of

the most popular heuristics is the Dollar Cost Averaging strategy: a Google search for this term

yielded no less than 34,700 sites.

3 Dollar Cost Averaging

Dollar Cost Averaging (DCA) is a strategy for the purchase of equity securities that is widely

recommended by professional investment advisors. The strategy calls for the periodic investment of

a fixed amount of money in a stock or portfolio each period over a given time interval, instead of

investing the whole amount in a lump sum at time zero. For example, an investor who wishes to

invest $10, 000 in stock XYZ may be advised to invest $1000 per month for ten months instead of

investing the whole $10, 000 at one time. The strategy may be applied on a stock by stock basis as

the previous example suggests, or it may be applied to investment in stocks as an asset class; for

example, an individual who receives an inheritance of $500, 000 may decide to invest $100, 000 per

year in stocks for each of five years, keeping the balance of his money in low risk money market or

bond accounts. The intuition behind Dollar Cost Averaging is that it allows the investor to purchase

stock at an average price which is below the average of the prices prevailing on the purchase dates

because the fixed periodic dollar investment purchases more shares when the price is low.

DCA is similar to other mechanical plans for investing such as portfolio rebalancing or the port-

folio insurance type of strategies which Brennan and Solanki (1981) and Leland (1980) have shown

to be optimal under certain circumstances. The major distinction between DCA and these strate-

gies is that DCA is a non-contingent strategy, whereas the hallmark of portfolio insurance and asset

allocation type strategies (including rebalancing to constant proportions) is that the investments are

contingent on the underlying stock price.

6
3.1 The Popular View

While discussions of DCA are now generally omitted from standard investments texts, presumably

because they are not thought worthy of serious discussion, a textbook discussion of DCA can be

found, for example, in Francis (1980) where it is described (page 706) as a ‘simple investment

plan which helps uninformed investors with the timing of their investments.’ The author continues

(page 707) that ‘one of the keys to earning high returns is to buy when the market is depressed.

However,.., amateur investors often sell at market lows. For people who lack the time or ability to

forecast market returns or who become upset during bear markets, the dollar averaging plan might

help them buy when the market is down and stocks are cheap - if they stick to the plan.’ It is

interesting to note that Francis’ account relates both to issues of self-control and to mean reversion

in stock prices, although neither of these phenomena was recognized by academics in finance at the

time the text was written. Similarly, Latane et al. (1975), page 517) describe DCA as a ‘technique

for taking advantage of stock market ups and downs.’ Another textbook of the same era written

by Cohen et al. (1977, page 48) also stresses the element of self-control: ‘the important thing is to

stick to your schedule..which, pyschologically, is usually hard to do.’ Even Sharpe’s (1978) classic

text concludes that ‘There is nothing inherently wrong with dollar averaging, but it has no magical

properties. Dollar averaging certainly does not change uncertainty per se from a vice to a virtue.’

(page 578)

Despite the fact that contemporary academic texts no longer contain any discussion of DCA, it is

still a strategy that is widely advocated in more popular publications. For example, Malkiel’s (1996)

Random Walk Down Wall Street, which is a popular distillation of academic advice on investment

tempered with some practical wisdom, has a section entitled Dollar-Cost Averaging can Reduce

the Risks of Investing in Stocks and Bonds. In it, he claims (page 356) that ‘This technique is

controversial, but it does help you avoid the risk of putting all of your money in the stock or bond

market at the wrong time.’ Interestingly, he does not explain why the technique is controversial, but

goes on to say that ‘(DCA) can reduce (but not avoid) the risks of equity investment by ensuring that

the entire portfolio of stocks will not be purchased at temporarily inflated prices’ and presents an

historical table which illustrates the ‘tremendous potential gains possible from consistently following

7
(DCA).’ Malkiel also stresses the psychological difficulty of actually following through on a DCA

strategy ‘no matter how pessimistic you are.’ It is clear that his rationale for the strategy is based

on a model of ‘market panics’ and ‘speculative explosions.’

3.2 The Academic View

In the pre-modern era there was significant interest among academics in ‘formula plans’ for investing

which included DCA as a special case. For example, the analyses of Ketchum (1947) and Weston

(1949) are predicated on the notion that stock prices are ‘cyclical’. Ketchum analyses the returns

to plans that alter the fund exposure to equity as a function of the level of the stock index relative

to its trend value (fitted ex-post) and, not surprisingly, finds that formula timing plans are superior

to 100% equity strategies, while Weston analyzes different formula plans in terms of the types of

price fluctuations which are implicitly being assumed. Cottle and Whitman (1951) discuss historical

simulations of formula plan returns, although they only present results for a constant bond-stock

mix. Not surprisingly, interest in formula plans waned as the random walk replaced cyclicality as

the standard model of stock price behavior.13

During the era in which the random walk theory was widely accepted as a good approximation

of stock market behavior, the few academic articles on DCA that appeared were uniformly negative.

Pye (1971), Neave and Wigginton (1976) and Constantinides (1979) criticize DCA on the grounds

that it makes the investor’s portfolio at any point in time dependent on its initial composition, so

that two investors with identical tastes and wealth who follow DCA strategies will hold different

portfolios at any point in time if the initial composition of their portfolios is different. But for an ex-

pected utility maximizing investor the optimal portfolio should depend only on current information.

Brennan and Solanki (1981) prove that DCA is suboptimal for an investor maximizing the expected

utility of terminal wealth when returns are iid. Rozeff (1994) demonstrates that, after adjusting

appropriately for risk within a mean variance framework, a lump sum investment policy is always

superior to a DCA policy if the stock market has a positive risk premium. However, his proof rests

13 The random walk hypothesis emerged in the papers of Kendall (1953), Roberts (1959), and Alexander (1961)

which are collected in Cootner’s (1964) classic The Random Character of Stock Market Prices.

8
on the assumption that successive market returns are independent. Dybvig (1988) presents a more

general approach for calculating the inefficiency cost of policies similar to DCA for arbitrary utility

functions, but again his analysis is predicated on the assumption of iid returns.

The random walk hypothesis is now generally rejected as an adequate description of stock price

behavior.14 Moreover, the psychological problems faced by an investor operating in highly uncertain

environment that are stressed in most popular accounts of DCA are now receiving increased attention

from academics.15 However, these developments have yet to spark a renewed academic interest in

formula plans, although Statman (1995) provides a behavioral interpetation of dollar cost averaging

and argues (page 75) that “the non-sequential rules of dollar-cost averaging reduce responsibility

and regret”.

In what follows we present empirical evidence on the performance of DCA strategies for an

investor whose objective is to maximize the expected value of a standard von Neumann-Morgenstern

utility function defined over wealth. Our emphasis is on the behavior of stock prices and their

implications for DCA. We ignore all issues that are related to modern developments in behavioral

finance, not because we think that such issues as self-control and confidence in the underlying model

are not important, but because they are too broad for us to introduce here.

4 Data

Security prices and firm market values are from CRSP for the period from December 1925 to De-

cember 2003. The market portfolio is taken as either the CRSP Equal Weighted Market Portfolio or

the CRSP Value Weighted Market Portfolio, and the riskless interest rate is the 30-day T-bill rate,

also taken from CRSP. A security is considered to be available for investment in a given month if the

security price is available at the end of the previous month. The total number of individual securities

is 25,396. The maximum number of securities available in any month is 9,265; the minimum is 513,

and the median is 2,143.

14 Summers (1986) was among the first to point out the weakness of the empirical foundations of the random walk;

De Bondt and Thaler (1985), Poterba and Summers (1988) and Jegadeesh and Titman (1993) among others present
evidence of ‘mean reversion’ in stock price behavior.
15 For a survey of this literature see Hirshleifer (2001).

9
 Stocks that are delisted without a delisting return being given by CRSP are assigned a return

of -100% if the delisting is due to liquidation or for performance reasons, or if the issue stopped

trading for unknown reasons;16 if the delisting return is due to a merger or exchange offer without

detailed information being available17 a delisting return of zero is assigned. After the delisting, the

proceeds (if any) are assumed to be invested in the risk free asset until the horizon and no further

investments in the risky asset are made.

5 Simulation

In order to assess the relative merits of the DCA strategy, Monte Carlo simulations were run for

purchases of the market portfolio and of individual securities. For each strategy s the expected

utility was calculated for a power utility function of the form:

Ws 1−γ
U (Ws ) = (1)
1−γ

for different values of γ, the coefficient of relative risk aversion. We analyze, in turn, DCA strategies

for investing in the market portfolio, for purchasing a single stock when no stocks are held initially,

and for purchasing a single stock when a portfolio of common stocks is already held.

5.1 Investing in the Market Portfolio

To evaluate the T month dollar cost averaging strategy of investing in the market portfolio, the

proceeds of an immediate lump-sum purchase of the market portfolio, and the proceeds of the dollar

cost averaging strategy, were calculated for investments of $1, starting in 10, 000 randomly chosen

months that were selected with equal probability between December 1925 and January 1998. For

each simulation, the wealth relatives for a buy and hold strategy of investing 100% of wealth in

the market portfolio for T = 12, 24, 36,48, 60, and 72 months were calculated. The corresponding

wealth relative for the T -month dollar cost averaging strategy was calculated by assuming that at

the beginning of the first month a fraction 1/T of the wealth was invested in the market portfolio

16 CRSP delisting code (DLSTCD) of 400-599 or 160.

17 DLSTCD of 200-399. Shumway (1997) reports average delisting returns of minus 30%.

10
and the balance in the risk free asset. Then, at the end of each month j = 1, · · · , T − 1, a fraction

1/(T − j) of the remaining wealth in the riskless asset was invested in the market portfolio. If the

riskless interest rate is equal to zero, this strategy implies that an additional fraction 1/T of the

initial wealth is invested in the risky asset each period. Thus the fraction of wealth left in the riskless
T −2
asset at the end of the first period is (1 − 1
T )(1 − 1
T −1 ) = T , and so on. This corresponds exactly

to what is meant by a dollar cost averaging strategy. With a positive interest rate, the fraction of

the initial wealth that is invested in the market increases at the interest rate. Let WisT denote the

wealth at the end of T months in simulation i from investing $1 under strategy s. The certainty

equivalent wealth for strategy s, CEs (T ) is defined as the amount of wealth to be received with

certainty at time T which would make the investor as well off as following strategy s. It is defined

by:
 10,000  1−γ
1

1  1−γ
CEs (T ) = WisT . (2)
10, 000
i=1

Panel A of Table 1 reports the certainty equivalents for four strategies for investing in the value-

weighted market portfolio for γ ranging from 2 to 7. The first is to buy and hold the market

portfolio (BH): under this strategy the whole of the initial dollar of wealth is assumed to be invested

in the market portfolio at time zero. The second is the dollar cost averaging strategy described

above (DCA). The third (50:50) is the strategy of investing 50% of the initial wealth in the market

portfolio at time zero and the balance in the riskless asset, and then rebalancing every month to

maintain the portfolio proportions. The fourth (OPT) is to invest the ‘optimal’ proportions in the

market and the risk free asset and to rebalance to maintain the proportions. The optimal allocation

to the market under the assumption of continuous rebalancing and iid returns, x∗, is given by:

(R̄M − Rf )
x∗ = (3)
γσM 2

where R̄M − Rf is the mean return on the market portfolio in excess of the riskless interest rate,

and σM is the volatility of the excess return. These parameters were estimated from the sample

data and were 7.4% and 19.1% for the value weighted market portfolio, and 11.5% and 26.1% for

the equal weighted market portfolio. Note that what we have called the optimal allocation will not

11
be truly optimal in that the rebalancing is at discrete intervals and market returns are not iid.

In fact, the certainty equivalent under the ‘optimal’ strategy does exceed that under the other

strategies in almost all cases. However, the optimal strategy is predicated on the assumption that

the investor knows, not only his risk aversion coefficient, but also the exact distribution of the

excess returns; it is therefore practically infeasible. The buy and hold strategy has the most variable

certainty equivalent across risk aversion levels; for the 36 month strategy, it falls from 1.26 for a risk

aversion of 2 to 0.48 for a risk aversion of 7. In contrast the DCA strategy is much more robust

across risk aversion levels, falling from 1.22 for RRA=2 to 0.87 for RRA equal 7. However, the

DCA strategy is uniformly dominated by the 50:50 strategy.18 It is interesting to note that the

expected wealth under DCA is almost identical to that under the 50:50 strategy: it follows that

the 50:50 strategy has less risk than the DCA strategy. These results are consistent with what one

would expect in a random walk market with no risk premium since then both the DCA and the

50:50 strategies would imply that on average 50% of wealth was invested in stocks. However, the

50:50 strategy achieves better time diversification since under this strategy exactly 50% of wealth

is invested in stocks each month, whereas under DCA the expected proportion of wealth in stocks

increases over time. Thus, there is no evidence that the claimed market timing advantage of DCA,

that it allows the investor to ‘avoid the risk of putting all of his money in the stock market at the

wrong time,’ is sufficient to offset the poor time-diversification that it causes.

Panel B reports corresponding results for the equal weighted market portfolio. The results are

similar. Thus there is no evidence that dollar cost averaging represents a good approach to investing

in the market portfolio.

5.2 Buying Individual Securities

Table 2 reports results that are similar to those reported in Table 1 except that, instead of investing

in the market portfolio, the investor is assumed to invest in a single security. The table presents the

certainty equivalents for two strategies: BH is the strategy of investing 100% of available funds in

18 We also considered a 50:50 buy and hold strategy which slightly outperformed the monthly rebalancing strategy
whose results are reported in Table 1.

12
the chosen security at the end of the randomly chosen month. DCA is the strategy of buying the

same security by dollar cost averaging as explained above. Panel A reports the results when the

security to be purchased is chosen from all the available securities with equal probability. Panel B

reports the results when the security is chosen with a probability that is proportional to the market

value of the security outstanding on the trade date. Returns on the two strategies are calculated in

the same manner as for the market purchases just described, allowing for the delisting of a security.

When the security is delisted, the proceeds that are calculated in the manner described in Section

4 are assumed to be invested in the risk free asset.

Three features are apparent from the figures in Table 2. First, most of the certainty equivalents

are less than unity, implying that investment in a single security leads to an economic loss even

without taking account of the time value of money represented by the interest rate. The reason for

this of course is the very poor diversification, and correspondingly low level of portfolio efficiency,

achieved with a single stock portfolio.19 Thus it is generally uneconomic to purchase a single stock

portfolio either outright or by a dollar cost averaging strategy. However, such a purchase may make

sense for a young person who is beginning to build a portfolio of individual stocks.20 Secondly, the

certainty equivalents tend to be significantly higher when the stock is chosen with a value weighted

probability. The reason for this is the lower level of total risk of large firms,21 and the correspondingly

higher level of portfolio efficiency they offer for a single stock portfolio. Thirdly, in all cases the

certainty equivalent of the DCA strategy exceeds that of the corresponding BH strategy. This is in

part due to the fact that the DCA strategy involves a lower average investment in the security and,

as we have seen, on average investment in a single risky security portfolio is uneconomic. However,

there appears to be more to it than this. The certainty equivalent almost always decreases with the

horizon for the BH strategy; but for DCA the certainty equivalent increases with the horizon for low

levels of risk aversion, and more so for value weighted security selection. For γ = 2 the certainty

equivalent grows with the horizon at a rate marginally above the riskless interest rate which averaged

4% during our sample period.

19 Brennan and Torous (1999) show that the certainty equivalent wealth from investing in a randomly selected

single-security portfolio for 10 years is $0.36.

20 It is known that a personal stock portfolio may have tax advantages over a mutual fund.
21 See Campbell et al. (2001).

13
5.3 A Portfolio Approach

The analysis of the previous section is limited by the fact that a single security portfolio is inherently

inefficient. In this section therefore we consider an investor who is adding a new security to an existing

portfolio. We explicitly consider adding only a single security to an existing portfolio, and compare

the value of doing this by way of a (marginal) lump sum purchase on the decision date (BH), and

doing it by periodic fractional purhcases under a DCA strategy. For many individual investors these

are the realistic alternatives. Moreover, to the extent that our measure shows that DCA dominates

the BH strategy on an individual security basis, it will also dominate for the purchase of (a marginal

amount of) of a portfolio of securities. We assume that the existing portfolio is the market portfolio.

While the assumption that the investor holds the value weighted portfolio is a natural one since it

flows from the logic of the capital asset pricing model, the gain from adding a new security to a

portfolio that is already well diversified is likely to be small. Therefore we shall also consider the

case in which the investor holds the equally weighted market portfolio, which in the US context

corresponds better to the case in which the investor’s initial portfolio is not well diversified.

Let W̃Te denote the wealth at the end of month T of an investor under his chosen or equilibrium

investment strategy e, and consider the marginal value of $1 invested in a strategy s whose payoff at

time T is a random variable, ỹsT . We denote this marginal value of strategy s at time T by M Vs (T ),

where

 
M Vs (T ) = E U  (W̃Te )ỹsT . (4)

Define the Marginal Value Ratio (MVR) of the DCA strategy relative to the BH strategy for a given

investment horizon T by:

M VDCA (T )
M V R(T ) = . (5)
M VBH (T )

If the investor who holds an existing portfolio with payoff W̃Te is indifferent between the DCA and

BH strategies then M V R(T ) = 1. M V R(T ) > 1(< 1) implies that the DCA strategy dominates (is

14
dominated by) the BH strategy.

We calculate the empirical counterpart to expression (5) for a T month horizon by randomly

selecting for each simulation i a starting month and a security to invest in. The compound value
e
of $1 invested in the market portfolio for the next T months, WiT , is calculated, along with the

payoff from investing in the chosen security under DCA and BH, which we denote by yDCA,i,T

and yBH,i,T . Then the realized marginal value for strategy s (s = DCA, BH) is calculated by

RM Vis = U  (W̃ieT )ỹisT , and the estimated marginal value ratio is calculated as:
10,000
 i=1 RM Vi,DCA
M V R(T ) = 10,000 . (6)
i=1 RM Vi,BH


Table 3 reports values of M V R(T ) for different values of the risk aversion coefficient. In Panel A

the securities are assumed to be selected with equal probability from those outstanding in the given

month. In Panel B the probability of selection is proportional to the firm market value. The figures

in bold are based on the assumption that the investor holds the value weighted market portfolio

and the figures below those on the assumption that the investor holds the equal weighted market

portfolio.

All of the figures in the table up to the 48 month horizon are greater than or equal to unity, and

most of them are considerably in excess of unity. For the 36 month horizon, the marginal value for

an investor who holds the market portfolio, of purchasing a random security by DCA exceeds that

of an immediate lump sum purchase by from 5% to 134%, depending on the assumption about risk

aversion, the weighting of the market portfolio held and whether the security is chosen with equal or

value weighted probability. We conclude that any investor who holds the (equal or value-weighted)

market portfolio would be better off following a dollar cost averaging strategy in purchasing a new

security than in purchasing the security outright. The advantage of DCA is greater when the security

to be purchased is selected with equal probability. This is consistent with smaller stocks having

returns that are more ‘anomalous’. Moreover, when the securities for purchase are selected with

equal probability the advantage of DCA is greatest for an investor who holds the equally weighted

market portfolio, and when the securities for purchase are selected with value weighted probability

the advantage of DCA is greatest for an investor who holds the value weighted market portfolio.

15
The relative advantage of the DCA strategy is maximized when it is executed over a 36-48 month

horizon.

We found in analyzing purchase strategies for the market portfolio that, while DCA dominated

BH, both strategies were dominated by a strategy of maintaining a constant 50% of wealth in the
e
market portfolio. Therefore Table 4 repeats the analysis of Table 3 but with WiT being determined

by a 50:50 mix of the riskless asset and the market portfolio rebalanced monthly. The effect of

reducing the riskiness of the investor’s pre-existing portfolio is to reduce the marginal value ratios,

which suggests that part of the (relative) marginal value of the dollar cost averaging strategy comes

from its marginal effect on reducing the risk of the pre-existing portfolio when that exceeds the

investor’s optimal level. The marginal value ratio for the 36 month strategy exceeds unity for all

values of relative risk aversion greater than or equal to 4. We suspect that few investors in practice

hold an optimally levered position in stocks, and the evidence in Tables 3 and 4 suggests that dollar

cost averaging remains good advice for a broad range of investors and, even when it is dominated

by the BH strategy, the margin is relatively small except for relative risk aversion of 2 where it can

be as high as 10% for the T = 36 months strategy when the portfolio leverage is 50%.

Finally, in order to determine whether the whole advantage of DCA in the purchase of individual

securities arises from its risk reducing effect on an initially excessively levered portfolio, the analysis

was repeated assuming that the investor’s initial portfolio is optimally levered. That is, the fraction

of the portfolio allocated to equities is assumed to be given by equation (3), and the portfolio is

rebalanced monthly to maintain the optimal weighting. This is an extremely conservative assumption

because we are assuming that the investor could have formed an optimal portfolio based on the

sample moments of portfolio returns over the whole period even though these were not available

until the end of the sample period. The results are reported in Table 5. The results for the 36

month DCA strategy may be summarized as follows. When an investor holds an optimal mix of the

value weighted market portfolio and the riskless asset, a 36 month DCA strategy for purchasing a

new security yields a marginal advantage of 3 − 6% over a traditional BH strategy. In contrast to

the results reported for non-optimally levered initial portfolios, the advantage of DCA appears to

be marginally greater when the securities for purchases are selected with value weighted probability;

16
this suggests that the effect is greatest for large firms. Curiously, the effect largely disappears, and

is even reversed for long horizons, when the investor is assumed to hold an optimally levered equal

weighted market portfolio.

The continuing advantage of the DCA strategy even when the initial portfolio is optimally levered

shows that its advantage does not arise from an inappropriately levered initial portfolio. The strategy

appears to have a fundamental benefit arising from unknown properties the autocorrelation and

lagged cross-correlation of returns.22 A general property of our results is that the advantage of the

DCA strategy for adding a new security to an existing portfolio is maximized when the strategy is

executed over a horizon of 36-48 months. In the following section we shall show that it is possible to

account for some of the characteristics of our DCA results by a model of stock price behavior which

allows the market price of a security to depart from its fundamental value; however, we are not able

to explain the maximum at 36-48 months.

6 A Model of Stock Price Behavior23

Following the general rejection of the random walk hypothesis, there is no longer a canonical model

of individual stock price behavior which can be used to explain the success of the DCA strategy

for adding a new stock to a portfolio. Moreover, no single model is likely to be appropriate for

securities of different size firms with different risk and information charactersitics. Nevertheless, to

gain some insight we shall extend the model which Poterba and Summers (1988) use to account for

mean reversion in stock prices, so that it allows also for some short run positive serial correlation

in returns. Thus, following Poterba and Summers, we write the stock price at time t, Pt , as the

product of the fundamental value, Pt∗ , and a stationary process, Ut :

Pt = Pt∗Ut (7)

22 Properties of the lagged cross-correlations of returns have been studied by Lo and MacKinlay (1990), Brennan et

al.(1993), Chordia and Swaminathan (2002).

23 We thank a referee for suggesting this analysis.

17
where ut ≡ lnUt follows the process:

ut = ρut−1 + σu (zu,t + η1 zu,t−1 + η2 zu,t−2 ) (8)

and the fundamental value, Pt∗, is assumed to satisfy the Capital Asset Pricing Model, so that the

fundamental return, R̃∗t ≡ (P̃t+1

∗
− Pt∗)/Pt∗ , satisfies:

R̃∗t = RF,t + β(R̃M,t − RF,t ) + t (9)

where zu,t and t are standard normal variables, E[zu,t ] = E[t ] = E[zu,t t ] = 0, and RM,t , RF,t are

the returns on the market portfolio and risk free asset at time t. The ARMA(1,2) model (8) allows for

positive correlation at short horizons. When η1 = η2 = 0 the model reduces to the ARMA(1,1) model

of returns used by Poterba and Summers. Market and stock returns are generated monthly using

equations (7),(8), and (9), using the parameter values given in Panel A Table 6. The parameters

RF , E[RM ], σM
2
, were set equal to the sample mean of risk free rate and the mean and variance of

the CRSP value weighted index over the period 1926-2003. The security specific parameters were

chosen to generate a pattern of positive serial correlation for short horizons followed by negative

serial correlation, and to correspond roughly to the pattern of marginal value ratios seen in Table

6. For the base case the security specific parameter values imply that the standard deviation of

the (log of the) market price to fundamentals ratio, Ũ , has a standard deviation of 38.4%.24 The

analysis of Section 5.3 whose results were reported in Table 5 was repeated, this time using 25,000

simulations for the market return and security specific variates, zu,t , t . First it was assumed that

the investor held an initial portfolio that was the optimal combination of the market portfolio and

the risk free asset, given his degree of risk aversion. Then, to simulate an initial portfolio that was

not fully diversified, it was assumed that his initial portfolio was the optimal combination of the risk

free asset and a portfolio whose return was equal to the simulated return on the market portfolio

plus a normally distributed return with mean zero and (annualized) standard deviation, σz equal to

5%. The marginal value ratios for various scenarios are reported in Panel B of Table 6. As before,

a marginal value ratio in excess of unity implies the superiority of the DCA strategy relative to

24 Black (1984) suggests that an efficient market is one in which ‘price is within a factor of 2 of value’.

18
BH. The figures in boldface relate to an investor who initially holds the optimal mix of the risk

free asset and the market portfolio; the figures below them to an investor who holds the less well

diversified portfolio. The results were found to be largely insensitive to the assumed value of relative

risk aversion, γ, and the table reports results only for γ = 2. Model 1 uses the base parameter values

and the other models use the base values except for the parameter noted.

Panel B of the table shows that the autocorrelations of returns are positive for lag 1 (0.06) and

then turn negative, peaking at -0.07 at lag 3. Campbell, Lo, and MacKinlay (1997) report auto-

correlations of 0.043 (0.171), -0.053 (-0.034),-0.013 (-0.033),-0.004 (-0.016) for lags 1-4 for monthly

returns on the CRSP value (equal) weighted portfolio for the period 1962-94. It is to be expected

that autocorrelations for portfolios will be more positive than for the individual securities because

of thin trading and liquidity effects.

Note first that, when σu = 0 so that the price is always equal to fundamental value and the return

therefore satisfies the CAPM, the investor is essentially indifferent between the DCA and BH. The

reason for this is that an investor holding an optimally levered market portfolio is indifferent at

the margin between an investment in the riskfree asset and a security that satisfies the CAPM.

When the investor holds the optimally levered imperfectly diversified portfolio the DCA strategy is

slightly inferior to BH but the difference is not economically significant except for very long execution

periods.

For the base case parameters, the advantage of DCA is of the order of 6% to 7% when the

execuion period is 4-5 years, which compares with the 7-9% reported in Table 5 for γ = 2. However,

while we were able to reproduce the magnitude of the gains found with actual returns, we were not

able to reproduce the attenuation in the gains at long horizons found with the actual returns. This

may be due to the restrictive nature of the pricing process that we have assumed which induces only

short run dependence in returns. It may also be due to the restriction in the simulations to only a

single process for all securities; as noted above, the process is likely to be different for securities that

differ in firm characteristics. For example, Brennan et al. (1993) have found that the prices of small

stocks tend to adjust more slowly to economy wide information, and Chordia and Swaminathan

19
(2000) have found that the returns on portfolios of high trading volume portfolios lead those of low

volume portfolios after adjusting for firm size.

As the parameters of the mispricing process were varied from the base case we found that an

increase (decrease) in the lag 1 autocorrelation correlation, induced for example by increasing (de-

creasing) ρ or η1 , tends to increase (decrease) the advantage of DCA, and increasing the lag 2

autocorrelation by increasing η2 also tends to increase the value of DCA. Of course, these general-

izations apply only for local variation in the parameters around the base case.

In summary, we have been able to reproduce the gross features of the DCA results in a model of

security mispricing that isbased on the Poterba-Summers model of transitory security mispricing.

7 Summary

In this paper we have analyzed the empirical properties of dollar cost averaging portfolio and security

purchase strategies. While such strategies are widely recommended by popular investment advisors,

they find no mention in the standard academic texts that are currently used to train MBA and

undergraduate students in finance. The only recent references that we have found to dollar cost

averaging in the scholarly literature emphasize the suboptimal nature of such strategies.

We have evaluated the usefulness of dollar cost averaging strategies by simulating a large number

of portfolio and security purchases over the period January 1926 to December 2003 and comparing

the certainty equivalent returns for an investor with power utility of two types of strategy. The first

strategy, which we call buy and hold (BH), purchases the security or portfolio in a randomly selected

month and holds it for a fixed T month period. The second strategy is the dollar cost averaging

strategy (DCA), in which equal amounts (in present value) are invested in the security in each of T

months.

When DCA is applied to purchases of the market portfolio, the results are heavily dependent

on the assumed risk aversion of the investor. DCA is superior to BH for more risk averse investors,

but only because such investors should not be investing the whole of their wealth in the market

portfolio. DCA is dominated by a BH strategy in which the investor invests a fraction of his wealth

20
in the market portfolio that is optimal given his risk aversion and the sample moments of the return

distribution. Of course, unlike the DCA strategy, the optimal strategy cannot be determined ex-ante

since it relies on estimates of the moments of the return distribution that are estimated from the

whole data sample. However, for all levels of risk aversion the DCA strategy is also dominated by a

strategy in which 50% of wealth is invested in the market portfolio and the portfolio is rebalanced

monthly to maintain the proportions. This result is broadly consistent with what we would expect

in a random walk market, for it is well known that an investor with power utility in such a market

should optimally maintain a constant fraction of his wealth in the market portfolio. The 50:50

strategy roughly approximates this.25 The DCA strategy has a strongly time varying allocation to

the risky asset that fails to achieve efficient ‘time-diversification’ so that the investor’s final wealth is

much more heavily dependent on the return in the last month of the period than it is on the return

in the first month.

When DCA is applied to the purchase of a single security it easily dominates a BH strategy,

but this is only because of the inefficiency of a single security portfolio. The advantage of the DCA

strategy lies in the fact that it is less heavily invested in this inefficient portfolio.

A more reasonable interpretation of the DCA prescription is that it should be applied to the

purchase of securities that are being added to portfolios that are already well diversified. We therefore

compute the marginal utility of the payoffs to simulated DCA and BH strategies for the investment

of $1 in a security and then calculate the ratio of the expected marginal utilities of the two strategies.

We find that the DCA strategies always dominate for an investor whose initial portfolio is the market

portfolio, either value weighted or equal weighted, for DCA execution periods of up to 48 months:

the advantage of a 3 year DCA strategy ranges from 5% to 134%, depending on the risk aversion of

the investor. In this sense DCA is a good strategy for the average investor. When the initial portfolio

is changed to a 50:50 mix of the market and the riskless asset, the advantage of DCA is mitigated;

however, the qualitative advantage of DCA is maintained, except for investors with very low risk

aversion (γ = 2 or 3). In order to determine whether the relative advantage of DCA arises from

25 Brennan and Torous (1999) show that modest departures from the optimal portfolio mix have only second order

effects on expected utility.

21
its ‘delevering’ effect on an initial portfolio that is over-levered relative to the investor’s assumed

risk aversion, the analysis was repeated assuming that the investor’s initial portfolio was optimally

levered, where the optimal leverage is determined using full sample data on portfolio returns. Despite

this ‘super-efficiency’ of the investor’s initial portfolio, DCA retains its advantage for all levels of

risk aversion considered when the underlying market portfolio is value-weighted; its advantage over

the standard BH strategy is of the order of 3 − 6% at the 36 month horizon where the advantage of

the DCA strategy is greatest. An interesting implication of the superiority of DCA for purchasing a

marginal stock is its inferiority for selling a stock. Liquidity considerations apart, individual stocks

should be sold in a block rather than liquidating a fixed number of dollars of stock each period.

We have shown using Monte Carlo simulation that the advantages of the DCA strategy for

purchasing a security to add to an existing portfolio could be attributable to transient security

mispricing of the kind proposed by Poterba and Summers (1988).

In summary, dollar cost averaging is a heuristic that has been almost entirely overlooked by

academics since the development of the random walk hypothesis, and which is suboptimal under

standard assumptions about capital markets. Despite this, the strategy appears to be useful in real

world capital markets. We have not attempted to consider the role of transactions costs or market

impact. Nor have we attempted to determine the characteristics of security price behavior that

account for the relative success of the strategy. These are topics for more formal investigations. As

mentioned above, our analysis considers the benefits of DCA only from the viewpoint of a ‘rational’

investor with a well-defined von Neumann-Morgenstern utility function. An important further line

of enquiry is to consider this and similar strategies from the viewpoint of an investor who is subject

to the behavioral biases that are currently receiving increased attention from academics.

22
8 References

Alexander, S.S., 1961, Price Movements in Speculative Markets: Trends or Random Walks, Industrial

Management Review, 2, 7-26.

Black, F., 1984, Noise, Journal of Finance, 41 , 529-543.

Blume, L., and D. Easley, 1992, Evolution and Market Behavior, Journal of Economic Theory,

58, 9-40.

Bodie, Z., R.C. Merton, and W. Samuelson, 1992, Labor Supply Fleibility and Portfolio Choice

in a Life-Cycle Model, Journal of Economic Dynamics and Control, 16, 427-449.

Brennan, M.J., and R. Solanki, 1981, Optimal Portfolio Insurance, Journal of Financial and

Quantitative Analysis, 3, 279-300.

Brennan, M.J., 2004, How Did it Happen?, Economic Notes, 3, 3-33.

Brennan, M.J., N. Jegadeesh, and B. Swaminathan, 1993, Investment Analysis and the Adjust-

ment of Stock Prices to Common Information, Review of Financial Studies, 6, 799-824.

Brennan, M.J., and [Link], 1999, Individual Decision Making and Investor Welfare, Eco-

nomic Notes, 28, 119-143.

Brennan, M.M., and Y. Xia, 2000, Stochastic Interest Rates and the Bond- Stock Mix,European

Finance Review, 4, 197-210.

Campbell, J.Y., A. Lo, and A.C. MacKinlay, 1997, The Econometrics of Financial Markets,

Princeton University Press, Princeton NJ.

Campbell, J.Y., and L. Viceira, Strategic Asset Allocation, Clarendon Press, Oxford, 2002.

Campbell, J.Y., M. Lettau, B. Malkiel, and Y. Xu, 2001, Have Individual Stocks become more

volatile?, Journal of Finance, 56, 1-43.

Canner, N., [Link], and [Link], 1997, An Asset Allocation Puzzle, American Economic

Review, 87, 181-191.

23
 Chordia, T., and B. Swaminathan, 2000, Trading Volume and Cross-autocorrelations in Stock

Returns,Journal of Finance, 55, 913-936 .

Cocco, J.F., F.J. Gomes, and P.J. Maenhout, 2003, Consumption and Portfolio Choice over the

Life-Cycle, Review of Financial Studies, forthcoming.

Cohen, Zinbarg, E.D., and A. Zeikel, 1977, J.B., Investment Analysis and Portfolio Management,

3rd. Edition, Irwin, Homewood.

Constantinides, G.M., 1979, A Note on the Suboptimality of Dollar-Cost Averaging as an In-

vestment Policy, Journal of Financial and Quantitative Analysis, 14 , 443-49.

Constantinides, G.M., 1983, Capital Market Equilibrium with Personal Taxes, Econometrica, 51,

611-636.

Cootner, P.H., 1964, The Random Character of Stock Market Prices, M.I.T. Press, Cambridge.

Cottle, C.S., and W.T. Whitman, 1951, Testing Formula Plans, Journal of Finance, 6, 220-228.

Dammon, R., C. Spatt, and H. Zhang, 2001, Optimal Consumption and Investment with Capital

Gains Tax, Review of Financial Studies, 14, 583-616.

De Bondt, W.F.M., and R. Thaler, 1985, Does the Stock Market Overreact?, Journal of Finance,

40, 793-805.

DeMiguel, V., L. Garlappi, and R. Uppal, 2005, How Inefficient is the Simple 1/N Asset-

Allocation Strategy?, unpublished manuscript, London Business School.

Dybvig, P.H., 1988, Inefficient Dynamic Portfolio Strategies or How to Throw Away a Million

Dollars in the Stock Market, Review of Financial Studies, 1, 67-88.

Francis, J.C., 1980, Investments, 3rd. Edition, McGraw-Hill, New York.

Hens, T., R. Amir, I. Evstigneev and Klaus Schenk-Hopp, Market Selection of Financial Trading

Strategies: Global Stability”, joint with Working Paper No. 83, Institute for Empirical Research in

Economics, University of Zurich forthcoming in Journal of Mathematical Economics, 2003.

24
 Hirshleifer, D., 2001, Investor Psychology and Asset Pricing, Journal of Finance, 1533-1598.

Jegadeesh, N., and S. Titman, 1993, Returns to Buying Winners and Selling Losers, Journal of

Finance, 48, 65-91.

Kendall, M.G., 1953, The Analysis of Economic Time Series –Part I: Prices, Journal of Royal

Statistical Association, 96, 11-25.

Ketchum, M.D., 1947, Investment Management through Formula Timing Plans, Journal of Busi-

ness, 20, 157-158.

Kurz, M.,1994a, On the Structure and Diversity of Rational Beliefs, Economic Theory 4, 877-900.

Kurz, M.,1994b, On Rational Belief Equilibria, Economic Theory 4, 859-876.

Latane, H.A., C.P. Jones, and D.L. Tuttle, 1975, Security Analysis and Portfolio Management,

2nd Edition, Ronald Press, New York.

Leland, H.E., 1980, Who Should Buy Portfolio Insurance?, Journal of Finance, 35, 581-594.

Lo, A., and C.A. MacKinlay, 1988, Stock Market Prices Do Not Follow Random Walks: Evidence

from a Simple Specification Test, Review of Financial Studies, 1, 41-66.

Lo, A., and C.A. MacKinlay, 1990, When are Contrarian Profits Due to Stock Market Over-

Reaction?, Review of Financial Studies, 3, 175-206.

Lo, A.,2002, Bubble, Rubble, Finance In Trouble?,Journal of Psychology and Financial Markets,

3, 76-86.

Malkiel, B.G., 1996, A Random Walk down Wall Street: The Best Investment Advice for the

New Century, W.W. Norton Co., New York.

Markowitz, H., 1959, Portfolio Selection: Efficient Diversification of Investments, Wiley, New

York.

Neave, E.H., and J.C. Wigginton, 1976, Dollar Averaging and Global Risk Aversion, School of

Business, Queens University, WP76-7.

25
 Odean, T., 1998, Are Investors Reluctant to Realize their Losses, Journal of Finance, 53, 1775-

1798.

Poterba, J., and L.H. Summers, 1988, Mean Reversion in Stock Prices: Evidence and Implica-

tions, Journal of Financial Economics, 22, 27-60.

Pye, G., 1971, Minimax Policies for Selling an Asset and Dollar Averaging, Management Science,

17, 379-393.

Roberts, H.V., 1959, Stock Market “Patterns” and Financial Analysis: Methodological Sugges-

tions, Journal of Finance, 14, 1-10.

Routledge, B.R., 1999, Adaptive Learning in Financial Markets, Review of Financial Studies 12,

1165-1202.

Routledge, B. R., 2001, Genetic Algorithm Learning to Choose and Use Information Macroeco-

nomic Dynamics 5, 2001, 303-325.

Rozeff, M., 1994, Lump-Sum Investing versus Dollar Cost Averaging, Journal of Portfolio Man-

agement, 20, 45-50.

Samuelson, P.A., 1969, Lifetime Portfolio Selection By Dynamic Stochastic Programming, The

Review of Economics and Statistics, 239-246.

Samuelson, P.A., 1989, The judgement of economic science on rational portfolio management:

indexing, timing, and , long-horizon effects, Journal of Portfolio Management, 25, 4-12.

Shumway, T., 1997, The Delisting Bias in CRSP, Journal of Finance, 53, 327-340.

Solomon, E., 1948, Are Formula Plans what they seem to be? Journal of Business, 21, 92-97.

Sharpe, W.F., 1963, a Simplified Model for Portfolio Analysis, Management Science, 9, 277-293.

Sharpe, W.F., 1978, Investments, Prentice-Hall, Englewood Cliffs.

Statman, M., 1995, A Behavioral Framework for Dollar-Cost Averaging, Journal of Portfolio

Management, 22, 70-78.

26
 Summers, L.H., 1986, Does the Stock Market Rationally Reflect Fundamental Values?, Journal

of Finance, 41,591-601.

Weston, J.F. 1949, Some Theoretical Aspects of Formula Timing Plans, Journal of Business, 22,

249-270.

27
Table 1

Certainty Equivalent (CE) Payoffs per Dollar Invested for Buy and Hold (BH), Dollar Cost Averaging (DCA), 50%
cash:50% stock (50:50), and optimal (OPT) strategies applied to the Value Weighted Market Portfolio (Panel A) and
Equal-Weighted Market Portfolio (Panel B) purchases starting in randomly chosen months between December 1925
and January 1998 and continued for up to 72 months.
The 50:50 strategy invests half the wealth in the riskless asset and half in the market portfolio and rebalances monthly. The
optimal strategy invests a fraction of wealth x∗ ≡ (R̄M − Rf )/(γ ∗ σM 2
) in the market portfolio and the balance in the risk
2
free asset, where R̄M − Rf is the mean excess return on the market portfolio, σM is the variance of the return on the market
portfolio. γ is the coefficient of relative risk aversion.

Panel A: Value Weighted Market Portfolio

Horizon (months)

Relative 12 24 36 48 60 72
Risk Aversion Strategy
2 BH 1.08 1.17 1.26 1.37 1.49 1.64
DCA 1.07 1.14 1.22 1.30 1.38 1.48
50:50 1.07 1.15 1.23 1.32 1.41 1.51
OPT 1.08 1.16 1.25 1.36 1.49 1.64
3 BH 1.05 1.07 1.09 1.19 1.30 1.44
DCA 1.06 1.12 1.18 1.25 1.31 1.40
50:50 1.07 1.13 1.20 1.28 1.37 1.47
OPT 1.07 1.13 1.19 1.28 1.38 1.50
4 BH 1.01 0.93 0.87 0.97 1.11 1.24
DCA 1.05 1.09 1.12 1.17 1.22 1.29
50:50 1.06 1.11 1.17 1.25 1.33 1.42
OPT 1.06 1.11 1.17 1.24 1.33 1.43
5 BH 0.96 0.78 0.69 0.79 0.95 1.07
DCA 1.04 1.04 1.04 1.08 1.11 1.15
50:50 1.05 1.09 1.13 1.20 1.29 1.38
OPT 1.05 1.10 1.15 1.22 1.29 1.38
6 BH 0.90 0.65 0.56 0.67 0.84 0.95
DCA 1.02 0.99 0.95 0.97 0.99 1.01
50:50 1.04 1.06 1.08 1.16 1.24 1.33
OPT 1.05 1.09 1.14 1.20 1.27 1.34
7 BH 0.83 0.56 0.48 0.58 0.75 0.86
DCA 1.00 0.93 0.87 0.87 0.88 0.89
50:50 1.03 1.02 1.02 1.10 1.20 1.28
OPT 1.05 1.09 1.13 1.19 1.25 1.31
Expected Wealth per BH 1.13 1.27 1.43 1.58 1.76 1.95
Dollar Invested DCA 1.09 1.18 1.28 1.37 1.48 1.60
50:50 1.08 1.18 1.28 1.38 1.49 1.62

28
 Panel B: Equal Weighted Market Portfolio
Horizon (months)

Relative 12 24 36 48 60 72
Risk Aversion Strategy
2 BH 1.09 1.19 1.30 1.44 1.63 1.86
DCA 1.08 1.17 1.26 1.36 1.48 1.63
50:50 1.08 1.18 1.28 1.39 1.52 1.67
OPT 1.09 1.19 1.31 1.45 1.63 1.85
3 BH 1.04 1.04 1.05 1.18 1.37 1.55
DCA 1.07 1.13 1.20 1.28 1.37 1.50
50:50 1.07 1.15 1.24 1.34 1.47 1.61
OPT 1.07 1.15 1.23 1.35 1.48 1.64
4 BH 0.99 0.87 0.78 0.89 1.12 1.25
DCA 1.05 1.08 1.12 1.17 1.23 1.32
50:50 1.06 1.12 1.19 1.29 1.41 1.55
OPT 1.06 1.13 1.20 1.30 1.41 1.54
5 BH 0.93 0.71 0.58 0.69 0.92 1.03
DCA 1.03 1.03 1.02 1.05 1.07 1.11
50:50 1.05 1.08 1.12 1.23 1.36 1.48
OPT 1.06 1.11 1.18 1.26 1.36 1.47
6 BH 0.86 0.59 0.47 0.56 0.77 0.87
DCA 1.01 0.97 0.92 0.92 0.92 0.94
50:50 1.04 1.04 1.05 1.16 1.30 1.42
OPT 1.05 1.10 1.16 1.24 1.32 1.42
7 BH 0.80 0.51 0.40 0.48 0.67 0.76
DCA 0.98 0.91 0.84 0.82 0.81 0.81
50:50 1.02 1.00 0.98 1.09 1.24 1.35
OPT 1.05 1.10 1.15 1.22 1.30 1.38
Expected Wealth per BH 1.18 1.39 1.60 1.87 2.14 2.42
Dollar Invested DCA 1.12 1.24 1.36 1.50 1.65 1.83
50:50 1.11 1.22 1.34 1.48 1.62 1.78

29
Table 2
Certainty Equivalent (CE) Payoffs for Buy and Hold (BH) and Dollar Cost Averaging (DCA) strategies ap-
plied to randomly chosen securities and months of purchase selected with equal probability between December 1925
and January 1998. Under DCA the purchases are continued for up to 72 months, and under BH the securities
purhcased are assumed to be held during the same period. Securities are chosen either with Equal Probability (Panel
A) or Value-Weighted Probability (Panel B).

Panel A: Equal Probability Security Selection

Horizon (months)
Relative 12 24 36 48 60 72
Risk Aversion Strategy
2 BH 0.90 0.56 0.43 0.38 0.36 0.35
DCA 1.02 1.04 1.09 1.13 1.15 1.20
3 BH 0.66 0.04 0.03 0.02 0.03 0.03
DCA 0.94 0.89 0.91 0.89 0.72 0.80
4 BH 0.38 0.01 0.01 0.01 0.01 0.01
DCA 0.79 0.60 0.67 0.57 0.31 0.37
5 BH 0.24 0.01 0.00 0.00 0.00 0.00
DCA 0.57 0.36 0.46 0.35 0.17 0.20
6 BH 0.17 0.00 0.00 0.00 0.00 0.00
DCA 0.41 0.24 0.33 0.24 0.12 0.13
7 BH 0.14 0.00 0.00 0.00 0.00 0.00
DCA 0.32 0.18 0.26 0.19 0.09 0.10

Panel B: Value Weighted Security Selection

Horizon (months)
Relative 12 24 36 48 60 72
Risk Aversion Strategy
2 BH 1.01 0.96 0.88 0.89 0.88 0.96
DCA 1.04 1.09 1.15 1.21 1.27 1.36
3 BH 0.91 0.46 0.09 0.10 0.10 0.10
DCA 1.02 1.02 1.06 1.11 1.12 1.20
4 BH 0.74 0.17 0.02 0.02 0.02 0.02
DCA 0.99 0.92 0.92 0.98 0.91 0.95
5 BH 0.54 0.09 0.01 0.01 0.01 0.01
DCA 0.94 0.76 0.75 0.82 0.68 0.69
6 BH 0.41 0.06 0.01 0.01 0.01 0.01
DCA 0.87 0.61 0.61 0.69 0.52 0.51
7 BH 0.33 0.05 0.00 0.00 0.00 0.00
DCA 0.78 0.50 0.50 0.59 0.41 0.40

30
Table 3
Marginal Value Ratios between Dollar Cost Averaging and Buy and Hold purchase strategy payoffs for an investor
who holds the market portfolio and commences purchases of randomly chosen securities in randomly chosen initial
months between December 1925 and January 1998 and continues the purchases (DCA) or holds the securities for up
to 72 months.
Securities are chosen either with Equal Probability (Panel A) or Value-Weighted Probability (Panel B). Figures in bold
font relate to the value-weighted market portfolio, while figures in standard font relate to an equally weighted market
portfolio. The Marginal Value Ratio is the ratio of the expected marginal utilities per dollar invested in the Dollar Cost
Averaging and Buy and Hold strategies.

Panel A: Equal Probability Security Selection

Horizon (months)
Relative 12 24 36 48 60 72
Risk Aversion
2 1.00 1.00 1.05 1.05 1.05 1.02
1.02 1.05 1.10 1.08 1.03 1.00
3 1.04 1.16 1.38 1.42 1.36 1.28
1.07 1.23 1.50 1.44 1.18 1.14
4 1.11 1.47 1.90 1.96 1.71 1.59
1.14 1.55 1.99 1.90 1.24 1.21
5 1.20 1.87 2.22 2.36 1.93 1.84
1.23 1.90 2.22 2.12 1.20 1.21
6 1.34 2.14 2.32 2.54 1.98 1.99
1.33 2.13 2.29 2.16 1.13 1.18
7 1.49 2.27 2.34 2.60 1.95 2.08
1.42 2.26 2.31 2.13 1.07 1.16

Panel B: Value Weighted Security Selection

Horizon (months)
Relative 12 24 36 48 60 72
Risk Aversion
2 1.01 1.03 1.06 1.06 1.04 1.02
1.01 1.04 1.05 1.02 0.96 0.93
3 1.04 1.16 1.31 1.32 1.25 1.21
1.05 1.18 1.31 1.23 1.03 0.97
4 1.09 1.38 1.66 1.69 1.48 1.40
1.10 1.39 1.61 1.48 1.08 0.94
5 1.17 1.59 1.87 1.96 1.62 1.54
1.17 1.59 1.75 1.59 1.10 0.87
6 1.28 1.70 1.95 2.08 1.68 1.61
1.24 1.70 1.80 1.59 1.10 0.81
7 1.39 1.75 1.99 2.13 1.67 1.64
1.31 1.75 1.83 1.55 1.11 0.76

31
Table 4
Marginal Value Ratios between Dollar Cost Averaging and Buy and Hold purchase strategy payoffs for an investor
who holds a 50:50 combination of the risk free asset and the market portfolio and commences purchases of randomly
chosen securities in randomly chosen initial months between December 1925 and January 1998 and continues the
purchases (DCA) or holds the securities (BH) for up to 72 months.
Securities are chosen either with Equal Probability (Panel A) or Value-Weighted Probability (Panel B). Figures in bold
font relate to the value-weighted market portfolio, while figures in standard font relate to an equally weighted market
portfolio. The Marginal Value Ratio is the ratio of the expected marginal utilities per dollar invested in the Dollar Cost
Averaging and Buy and Hold strategies.

Panel A: Equal Probability Security Selection

Horizon (months)
Relative 12 24 36 48 60 72
Risk Aversion
2 0.97 0.93 0.90 0.87 0.86 0.85
0.98 0.95 0.92 0.89 0.87 0.85
3 0.99 0.96 0.95 0.93 0.93 0.91
1.00 0.98 0.98 0.95 0.92 0.89
4 1.00 1.00 1.02 1.02 1.02 0.99
1.02 1.03 1.06 1.03 0.98 0.94
5 1.02 1.05 1.14 1.15 1.14 1.09
1.04 1.10 1.20 1.14 1.03 1.00
6 1.04 1.14 1.32 1.32 1.28 1.22
1.06 1.20 1.41 1.31 1.09 1.05
7 1.06 1.26 1.55 1.55 1.43 1.35
1.09 1.33 1.67 1.51 1.13 1.09

Panel B: Value Weighted Security Selection

Horizon (months)
Relative 12 24 36 48 60 72
Risk Aversion
2 0.98 0.96 0.94 0.92 0.91 0.89
0.98 0.96 0.94 0.91 0.88 0.86
3 0.99 0.99 0.98 0.97 0.95 0.94
1.00 0.99 0.97 0.94 0.90 0.88
4 1.00 1.02 1.04 1.03 1.01 1.00
1.01 1.03 1.03 0.98 0.93 0.90
5 1.02 1.07 1.13 1.12 1.09 1.07
1.03 1.08 1.12 1.05 0.95 0.91
6 1.04 1.14 1.26 1.25 1.19 1.16
1.04 1.15 1.25 1.14 0.98 0.92
7 1.06 1.23 1.42 1.40 1.29 1.25
1.07 1.25 1.41 1.25 1.01 0.91

32
Table 5

Marginal Value Ratios between Dollar Cost Averaging and Buy and Hold purchase strategy payoffs for an investor who
holds the optimal combination of the risk free asset and the market portfolio, and commences purchases of randomly
chosen securities in randomly chosen initial months between December 1925 and January 1998 and continues the
purchases (DCA) or holds the securities for up to 72 months.
Securities are chosen either with Equal Probability (Panel A) or Value-Weighted Probability (Panel B). Figures in bold
font relate to the value-weighted market portfolio, while figures in standard font relate to an equally weighted market
portfolio. x∗ ≡ (R̄M − Rf )/γσ2 M is the optimal allocation to the market portfolio for a myopic investor, where γ is the
coefficient of relative risk aversion, σM is the sample volatility of the market portfolio, and R̄M − Rf is the sample excess
return on the market portfolio. The Marginal Value Ratio is the ratio of the expected marginal utilities per dollar invested
in the Dollar Cost Averaging and Buy and Hold strategies.

Panel A: Equal Probability Security Selection

Horizon (months)
Relative x∗ 12 24 36 48 60 72
Risk Aversion
2 1.06 1.01 1.02 1.07 1.09 1.08 1.05
0.89 1.01 1.02 1.04 1.02 0.98 0.95
3 0.71 1.00 1.01 1.05 1.05 1.05 1.02
0.59 1.01 1.01 1.03 1.00 0.96 0.93
4 0.53 1.00 1.01 1.05 1.04 1.04 1.01
0.45 1.01 1.01 1.02 0.99 0.95 0.91
5 0.42 1.00 1.01 1.04 1.04 1.04 1.01
0.36 1.01 1.01 1.02 0.98 0.94 0.90
6 0.35 1.00 1.01 1.04 1.04 1.04 1.01
0.30 1.01 1.01 1.02 0.98 0.93 0.90
7 0.30 1.00 1.01 1.05 1.04 1.04 1.01
0.25 1.01 1.01 1.01 0.97 0.93 0.89

Panel B: Value Weighted Security Selection

Horizon (months)
Relative x∗ 12 24 36 48 60 72
Risk Aversion
2 1.06 1.01 1.04 1.08 1.09 1.07 1.05
0.89 1.01 1.02 1.02 0.98 0.94 0.91
3 0.71 1.01 1.03 1.06 1.06 1.04 1.02
0.59 1.00 1.01 1.00 0.97 0.92 0.89
4 0.53 1.01 1.03 1.06 1.05 1.03 1.01
0.45 1.00 1.01 1.00 0.96 0.91 0.88
5 0.42 1.01 1.03 1.05 1.05 1.03 1.01
0.36 1.00 1.01 1.00 0.95 0.91 0.88
6 0.35 1.01 1.03 1.05 1.04 1.03 1.01
0.30 1.00 1.01 0.99 0.95 0.90 0.87
7 0.30 1.01 1.03 1.05 1.04 1.03 1.01
0.25 1.00 1.01 0.99 0.95 0.90 0.87

33
 Table 6

Marginal Value Ratios between Dollar Cost Averaging and Buy and Hold purchase strategy payoffs for an investor
who holds the optimal combination of the risk free asset and the market portfolio for simulated market and security
returns. Security and Market Returns are simulated using equations (7),(8), and(9). The Marginal Value Ratio is the ratio
of the expected marginal utilities per dollar invested in the Dollar Cost Averaging and Buy and Hold strategies. Ratios shown
in boldface are for an investor who holds an optimally levered position in the market portfolio; ratios shown in standard type
are for an investor who holds an optimally levered position in a portfolio with a β of unity and the same expected return
as the market portfolio, but with a residual risk of 5% per year. The parameter values used for Panel C are the base case
parameters excepts as noted.

Panel A: Base Case Parameter Values

Rf = 0.037 E[RM ] = 0.114 σM = 0.191 σz = 0.05
β = 1.0 σ = 0.05 σzu = 0.69 ρ = 0.85 η1 = 0.15 η2 = 0.05
γ=2

Panel B: Base Case Autocorrelations

Lag (months)
1 2 3 4 5 6 7 8 9 10 11 12
Autocorrelation 0.06 -0.04 -0.07 -0.06 -0.05 -0.05 -0.04 -0.03 -0.03 -0.03 -0.02 -0.02

Panel C: Marginal Value Ratios

Horizon (months)
Model Parameters 12 24 36 48 60 72

1. Base 0.999 1.035 1.061 1.076 1.074 1.081

0.997 1.030 1.053 1.065 1.061 1.066
2. σu = 0.0 1.001 1.000 1.001 1.003 0.997 1.000
0.999 0.996 0.994 0.994 0.985 0.985
3. β = 0.0 0.000 1.037 1.060 1.071 1.073 1.080
1.000 1.037 1.059 1.070 1.073 1.081
4. ρ = 0.900 0.976 1.020 1.057 1.076 1.091 1.103
0.974 1.015 1.050 1.067 1.077 1.087
5. η1 = 0.25 0.990 1.044 1.068 1.080 1.085 1.100
0.988 1.039 1.060 1.070 1.071 1.083
6. η2 = 0.150 0.997 1.039 1.060 1.074 1.088 1.089
0.995 1.035 1.053 1.065 1.074 1.074

34

View publication stats