Working Paper No. 101
THE CARLO ALBERTO NOTEBOOKS
Liquidity and Competition in Unregulated Markets
LIQUIDITY AND COMPETITION IN UNREGULATED MARKETS
Despite reputedly widespread market manipulation and insider trading, we find surprisingly
high liquidity and low transactions costs for actively traded securities on the NYSE between
1890 and 1910, decades before SEC regulation. Moreover, market makers behave largely as
predicted in theory: stocks with liquid markets and competitive market makers (cross-trading at
the rival Consolidated Exchange) trade with substantially lower quoted bid-ask spreads and with
less anti-competitive behavior (price discreteness). Effective spreads, illiquidity, and volume all
improve monotonically over time. Notably, the asymmetric information component of effective
spreads increases in relative and absolute terms from 1900 to 1910.
* Johns Hopkins University. http://www.econ.jhu.edu/people/fohlin/index.html
† University of Freiburg and CEPR
‡ University of Freiburg
We are grateful to Richard Sylla for help and advice on the institutional details of the NYSE as well as to
participants at the BETA Workshop at Strasbourg (May 2007), the Collegio Carlo Alberto in Moncalieri, Humboldt
University, and the ASSA meetings in New Orleans (January 2008). We thank particularly Giuseppe Bertola,
Elroy Dimson, John Latting, Monika Merz, Giovanna Nicodano, Albrecht Ritschl, Filippo Taddei, Harald Uhlig,
Andrey Ukhov, and Marc Weidenmier. We are indebted to Seth Friedman, Purvi Maniar, Michael Mueller, and
Frederic Rey for unflagging research assistance in gathering the daily stock data and to the U.S. National Science
Foundation (grants SES0137937 and SES0331009 to Fohlin) for financial support. Gehrig gratefully
acknowledges the hospitality of the Collegio Carlo Alberto in Moncalieri.
Copyright 2008 by Caroline Fohlin, Thomas Gehrig and Tobias Brünner. Any opinions expressed here are those of
the author and not those of the Collegio Carlo Alberto.
LIQUIDITY AND COMPETITION IN UNREGULATED MARKETS
Financial crises nearly always generate increased government involvement in the regulation of
financial markets and institutions. The collapse of financial markets and myriad banks between
1929 and 1933, followed as it was by a slew of new federal regulation, remains the most
significant example of the crisis-response pattern in the United States. While most would agree
that resulting institutions like the Securities and Exchange Commission (SEC) have provided
useful oversight and regulatory functions over the past 70 plus years, we actually know very little
about the functioning of US financial markets prior to the onset of government regulation.
Thus, in this paper, we examine liquidity provision and transactions costs—key
parameters of market performance—in the New York Stock Exchange between 1890 and 1910.
Using a newly-gathered database of daily closing prices, quoted bid and ask prices and trading
volume for all stocks traded on the exchange in those years, we calculate various measures of
illiquidity and transactions costs, in order to estimate the quoted and effective costs of trading.
In order to understand the sources of trading costs and market liquidity, we decompose the
effective spread measures into asymmetric information and order-processing components, based
on theoretical models of spread components. We then analyze the cross-sectional determinants
of the quoted spreads and market liquidity, including the impact of simultaneous trading on the
main competing exchange, the Consolidated Exchange.
Given the relatively early phase of development of equity trading in the NYSE, along
with the rudimentary communications technology of the pre-WWI era, we expect to find high
spreads by modern U.S. standards. Moreover, given the absence of regulation regarding insider
trading, one might be tempted to expect a relatively high adverse selection component in quoted
spreads and illiquid trading. At the same time, however, we expect that parallel trade on the
Consolidated likely increased competition (even if it also decreased efficiency via market
fragmentation) and therefore tightened spreads on the NYSE, particularly for stocks traded on
In fact, some of our hypotheses are borne out by the data, but there are many surprises.
Indeed, for the most heavily traded stocks, trading costs are roughly comparable to those in
developed markets at the end of the century. So we definitely have to revise our expectation that
markets operated with dramatically higher trading costs at this time. Likewise we find that
measures of market illiquidity for the highest volume stocks are comparable to the same measure
for stocks traded at the NYSE at the end of the 20th century. In the cross sectional analysis, we
confirm our hypothesis that stocks with liquid markets—those with a high number of trading
days and large volume—trade at lower cost, and those with simultaneous Consolidated activity
trade with significantly lower spreads. Over time, however, quoted spreads do not behave in line
with other measures of market liquidity and trading costs: while quoted spreads increase from
1900 to 1910, effective transaction costs and market illiquidity decrease monotonically and
trading volume of common stocks increases monotonically from 1890 to 1910. Preferred shares
show a different volume pattern, with the more actively traded shares increasing in volume from
1890 to 1900 and then dropping off again by 1910. Also of note, the asymmetric information
component of trading costs increases in relative and absolute terms from 1900 to 1910.
This study contributes to a growing line of research into the microstructure and
performance of securities markets prior to government regulation. Few other works have
investigated the historical development of transactions costs and liquidity in U.S. markets,
particularly the NYSE. To our knowledge, the current paper is the first to study these
phenomena for the NYSE using high-frequency historical data. Most closely related to our work
is Jones (2002), which examines month-end quoted bid-ask spreads for the components of the
Dow Jones Industrial average between 1900 and 2000. Jones (2002) shows that transaction
costs—he considers bid-ask spreads and commissions—explain a small part of the equity
premium over this long period. Also, spreads seem to be good predictors for future returns. The
purpose of that study is to follow very long-run movements in transactions costs. Due to the
lower frequency of the data and the smaller sample (of mostly the largest firms), the results
cannot readily be compared with those presented in the current paper. Moreover, our results
suggest that quoted bid-ask spreads are not a good measure for comparing the relevant
transaction costs over time. Rather measures of effective transaction costs or market illiquidity
are more reliable.
Also related to our work, Brown et al. (forthcoming) argue that direct competition with
the Consolidated Stock Exchange between 1885 and 1926 reduced transaction costs at the
NYSE. While they identify the competition effect around structural breaks in the time domain,
our cross-sectional regressions measure the competitive impact of firms being traded both at the
NYSE and the Consolidated Stock Exchange. In this sense we can also measure the value of a
“cross-listing” on both exchanges. As with Jones (2002), Brown et al. (forthcoming) rely on one
trading day per year over an extended period for a small sample of stocks, as opposed to our
daily data on the complete set of traded stocks. Additionally, we find that price clustering at
whole and half dollar increments – a potential indication for non-competitive conduct – is less
pronounced for securities traded on both exchanges both in 1900 and 1910.
Davis et al. (2007), focusing on capacity constraints and their softening after the seat sale
of 1928, also produce quoted bid-ask spreads for a selected high and low volume days
surrounding the sale. Mean spreads range from .65 to 1.69 percent: far lower than the estimates
we produce for the full set of stocks traded in 1900/1910 but similar to the spreads on high
volume stocks in these years. Their cross-sectional results, however, produce similar predictive
factors as we find for the earlier period. Their study also has data insufficient to create the more
extensive measures of effective transactions costs and market illiquidity that we use.
More similar from a methodological standpoint, Gehrig and Fohlin (2006) study trading
costs in the Berlin Stock Exchange, using daily prices, for a similar time period (1880-1910).
Due to the nature of trading in that market, however, the data are also quite different from the
NYSE data: the Berlin market produced only one daily price quote and no reported bid-ask
spreads. The results of that study indicate that estimated effective spreads in Berlin ranged
between approximately 11 and 28 basis points, while round-trip transactions costs varied from 45
to 116 basis points. In both cases, the measures declined over time, but unlike the New York
Exchange they were already fairly low by historical standards and certainly a lot lower than the
average at the New York Stock Exchange at that time.
These findings are particularly interesting in comparison with recent developing markets
(Lesmond, 2005). Our estimates indicate that NYSE illiquidity at the turn of the 20th century
was roughly comparable to emerging stock markets of China, the Czech Republic and Mexico at
the end of the 20th century.
The rest of the paper is organized as follows: the next section reviews the historical
context of the New York Stock Exchange and key features of corporate finance practice in the
pre-World War I era. Section III describes the theoretical underpinnings of various measures of
transactions costs, while section IV introduces the newly created database on daily stock prices,
volumes, and spreads. Section V presents the quoted and estimated effective spreads and their
decompositions, and section VI investigates the cross-sectional determinants of quoted spreads.
The final section concludes.
II. The Development of the NYSE before World War I
The New York Stock Exchange was created in 1792 when twenty four brokers and merchants
signed the Buttonwood Agreement. At this time, five securities were traded on the exchange
(nyse.com).4 In the first half of the nineteenth century, government issues comprised the bulk of
publicly traded securities. By 1860, the liberalization of incorporation law (Hickson and Turner,
2005) allowed for the creation of marketable securities to trade on organized exchanges. Within
a few years, railroads began issuing securities for trading on the large public markets in order to
satisfy their growing demands for capital. Around the 1880s, rail stocks made up a substantial
majority of the trading on the NYSE. Listings and trading on the exchange grew rapidly, and the
mix of securities changed, in the latter part of the nineteenth century and the first decades of the
twentieth. While railroad securities remained important, they lost some market share around the
turn of the twentieth century, as other sectors expanded more rapidly.5 By 1910, “non-rails”
outnumbered railroads on the NYSE for the first time since 1870 (Davis and Cull 1994). The
value of securities listed on the exchange exceeded $26 billion (about $500 billion in 2005
Rules and Regulation:
In the period of our study, the NYSE was owned by its members and largely self-regulated.
Among the key regulations were those dealing with membership. Joining the exchange was a
costly venture: a new member had to pay a membership fee and then buy the seat of an existing
4 See Michie (1986), p. 173 and Mulherin et al. (1991), p. 597, for surveys of the institutional development of the
NYSE. See also Baskin (1988).
5 Navin and Sears (1955) consider the 1880s the beginning of the shift to the widely-held industrial enterprise,
particularly due to the “trust” movement in the processing industries. The main trusts created in the 1880s involved
oil refining (Standard Oil), cotton oil refining, linseed oil refining, whiskey distilling, sugar refining, and lead
smelting and refining. They also credit the heavy demand for trading in trust certificates with New York’s rise to
preeminent exchange during the late 1880s.
6 Davis and Neal (1998) report a figure of 5.4 billion pounds, based on Michie (1987). The dollar values come from
using the calculator provided by www.measuringworth.com.
member. The exchange had fixed the number of seats at 1,100 in 1879, so that the prices of seats
varied with the market. These prices ranged between $4,000 and $4,500 in 1870 (approximately
$65,000 in 2005 values) and between $64,000 and $94,000 in 1910 (roughly $1-2 million in
The Governing Committee of the exchange held ultimate responsibility for exchange
operations and had the power to fine or even expel members for infractions against exchange
rules. The value of a member’s seat worked as collateral in these cases or in the event of
bankruptcy (Mulherin et al., 1991, 597-598). The courts upheld these powers as well as the
exchanges’ right to restrict trading solely to its members and to set other rules (Mulherin et al.,
The NYSE implemented relatively stringent listing standards and requirements, including
registration of all shares (to prevent stock watering), minimum shareholder numbers, and
qualitative assessment of risk. Oil stocks, for example, could not be listed in their early years, as
they were deemed too risky.
External regulation of exchange operations or of listed corporations came much later, and
corporate reporting law generally remained weak in the United States up until the Great
Depression. Private incentives, particularly the desire to access outside funds from investors,
encouraged more and more firms to disclose their balance sheets and income statements. In
1895, the NYSE began recommending that listed companies provide both a balance sheet and an
income statement in annual reports to investors. Such reporting became mandatory in 1899
(nyse.com).8 The content of these reports varied significantly in their breadth and accuracy, and
accounting standards and auditing practices took many more decades to evolve into what would
become the modern norm.
7 Michie (1986, p. 175), presumably reported in nominal terms.
8 See as well Archambault and Archambault (2005), who find that even as of 1915, listed companies—particularly
industrials that were not already regulated by the government—were significantly more likely to report an income
statement than unlisted companies. Similarly, listed companies were far more likely to report a balance sheet.
Organization of trading:
Though it started out operations using a call auction system, the NYSE moved to a continuous
auction method in 1871.9 Under this system, transactions occurred throughout the trading day at
whatever terms could be agreed upon by the parties involved, with no guarantee of a single
price.10 While the continuous auction method eliminated the problem of overcrowding and the
excessive time taken in the call auction, it created new problems of order imbalance—the
brokers interested in trading a given security may not arrive simultaneously at the particular
trading post for that security. In general, such random arrival reduces market liquidity, creating
greater order imbalance and price volatility compared to a call auction (Kregel, 1995).
The evolution of the trading method led to the creation of two distinct types of
intermediaries. The first type, brokers, traded on behalf of their customers and received set
commissions as their payment. The others, jobbers, bought and sold shares in order to make
markets in securities, and they received the spread between bid and ask prices as their
compensation. The increasing number and sophistication of jobbers then encouraged their
specialization in particular stocks, hence the term ‘specialist.’ These specialists made a market
in their stocks at a single trading post, and they traded on their own account as well as on behalf
of their customers.
Competition from other Exchanges:
The NYSE’s restrictive membership and listing rules led to the repeated rise of competitors from
its inception.11 The most significant competition came with the creation of the Consolidated
9 Kregel (1995, p. 464) gives a number of reasons relating to inefficiencies of the call auction. Kregel finds
unconvincing Garbade and Silber’s (1979) explanation for the shift to continuous trading—that the Civil War
increased the arrival of new information to the market.
10 See Kregel (1995) on the evolution of securities market organization in London and New York.
11 See, for example, Garvey (1944).
Stock Exchange, formed out of the merger of several rival exchanges, in 1885. The
Consolidated included 2,403 members—more than double the number of NYSE members—and
many brokers also traded in the unorganized “curb” market. By 1908, the three exchanges
contained 424 million shares of stock, over half of which (53.5 percent) were traded outside of
the NYSE (Michie, 1986, 175-176).
Compounding the incentives to deal outside of the NYSE, the brokers of the New York
Stock Exchange charged a fixed minimum commission of 1/8 percent on trades.12 The
Consolidated Stock Exchange, by contrast, charged a commission rate of 1/16, thus encouraging
nonmembers of the NYSE to deal on the Consolidated using NYSE market prices (Michie, 1986,
p. 178). By using the NYSE quotes, brokers of the Consolidated Exchange saved on the costs of
creating a price discovery mechanism, and were thereby able to charge lower commissions than
the NYSE (Mulherin et al., 1991, 608).
The NYSE worked continuously but not altogether successfully to eliminate its
competition. It created an Unlisted Trading Department to trade in stocks of the Consolidated
(Mulherin et al., 1991, 609), tried to remove tickers from the Consolidated Stock Exchange and
from outside brokers, and later forbade phone links to the Consolidated Stock Exchange. The
latter efforts failed, however, because brokers with legitimate access to the NYSE would trade at
the Consolidated at NYSE prices (Michie, 1986, 178).13 In 1896, dealing in differences between
domestic exchanges was banned and in 1898 the exchange banned the transmission of
continuous price quotes (Michie, 1986, 179).
The anti-competitive measures proved difficult to enforce, but they still limited
transactions between the NYSE and other domestic exchanges and created price differentials.
12 A loophole in the rule, however, allowed commissions of 1/32 percent (and often as low as 1/50 percent) on trades
for members buying and selling from each other. This discount pertained to all partners of a member firm, and thus
fostered the growth of large brokerage firms (Michie, 1986, 177-178). The original Buttonwood agreement stated a
minimum commission of ¼ percent.
13 For a detailed description of the legal battle for exchanges to control their quotes see Mulherin et al., 1991
The restrictive rules of the NYSE therefore limited the market in some securities, but
simultaneously hindered access to current prices by traders in the Consolidated. As Michie
(1986) points out, the “New York Stock Exchange covered only part of the New York market
and prevented the remainder from operating as efficiently as possible.”
Transactions Costs: Information and Competition
Trading in securities brought with it numerous costs, relating to both information asymmetries
and order processing. Information discrepancies between insiders and outsiders raised costs that
could be only partially offset by corporate reporting. The use of continuous trading created
additional illiquidity risk, particularly in stocks with thin markets that often required specialists
to hold inventories in order to make deals. Specialists required compensation for bearing these
risks, and the resulting spreads added to overall transactions costs. Moreover, restricted
memberships, minimum commission rates for brokers, and specialization in securities
(effectively product differentiation) may have lessened competition and allowed some market
power in the setting of spreads by specialists.
On the positive side, the innovations of the telegraph, ticker, and telephone lowered the
costs associated with disseminating information and expanded the NYSE’s geographical reach
(Mulherin et al., 1991, 606). It also allowed competing markets to gain access to NYSE quotes
and facilitated competition between the NYSE and other exchanges. These effects should have
helped lower order processing costs. To the extent that the exchange limited the listing of issues
judged to be too risky, the resulting selection bias should have mitigated the asymmetric
information components of transactions costs.
III. Measuring Market Liquidity and Transactions Costs
The development of the microstructure of the New York Stock Exchange prior to World War I
provides a unique real world experiment on the evolution of trading systems and associated
market liquidity and transactions costs in an unregulated (or self-regulated) environment.
Trading costs further reflect information about information asymmetries and market power.
Together, these various measures, and their underlying explanatory factors, provide a wide-
ranging picture of market functioning.
It is useful first to differentiate quoted spreads from effective spreads. While quoted
spreads can be readily observed as raw data, effective spreads need to be estimated by statistical
methods. Typically, effective spreads are more informative about real trading costs, because
quoted prices often change soon after a transaction has taken place or because traders can
actually negotiate to trade at prices between the quotes. When quotes change frequently during
the day, the effective costs of a round-trip transaction is likely to be lower than the quoted spread
at any point in time, because a hypothetical trader could take advantage of the option value of
trading the second part of a round-trip transaction at a different more favorable points in time
later. Also price improvements in form of bilateral agreements between traders and market
makers are rather customary in so-called quote drive trading systems, (i.e. in trading systems
based on market makers.)
Depending on the availability of empirical observations, estimated spreads can be
decomposed into the various theoretical components, like information, inventory holding, and
order processing cost. Since this decomposition of the underlying cost components is largely
based on theoretical considerations, in the sequel we briefly outline the theoretical basis for the
subsequent empirical analysis.
Market makers in an asset market receive as their compensation the difference between the price
paid to sellers and the price obtained from buyers—the bid-ask spread. Empirically, the
difference between quoted ask and bid prices, normalized by the midpoint of bid and ask prices
of the asset, provides an estimate of the actual transaction cost. Transactions do not necessarily
take place at quoted bid and ask prices, however, meaning that quoted spreads are not necessarily
precise reflections of real transactions costs. Moreover, the quoted spread wraps up a range of
different transactions costs: order processing expenses, inventory risk, asymmetric information,
and potentially monopoly rents. A number of alternative methods have been devised to more
accurately depict transactions costs and to allow decomposition of the spread into various
Realized Spreads and their Components
In order to estimate realized spreads, we use the method proposed by George et al. (1991). This
method refines and extends the serial covariance measure proposed by Roll (1984). In that
denoting the transactions return on a security i in period t, where
is the natural logarithm of the price of stock i, the Roll measure
estimate of security i’s effective spread.14 Since trades often take place at prices between the
quoted bid and ask prices, the estimated effective spread is smaller than a quoted spread. The
underlying idea of this estimator is that, in informationally efficient and stationary markets,
variation in transactions prices results from the randomness of buy and sell orders plus positive
transaction costs. In liquid markets with low transaction costs, successive individual orders have
little impact on observed transaction prices. In thin markets, price effects of individual trades
14 The transactions return is based on observed transactions prices. Transactions returns typically differ from true
returns, because even in efficient markets transactions costs prevent arbitrage, when true returns and transactions
returns are close enough.
may be more pronounced. If transaction costs are higher, the deviation of transaction prices from
true fundamentals will not be immediately arbitraged, even in efficient markets. Therefore, the
covariance of successive price changes provides information about market liquidity, and hence,
effective transaction costs.15 In liquid markets the covariance of successive prices will be low as
long as the price changes are not caused by systematic factors such as new market information.
And even new information will be reflected in prices immediately. In less liquid markets the
covariance will be higher, both, because of a larger market impact of individual trades, and
because information revelation is slower. The effective spread therefore arguably offers a better
estimate of actual transaction costs than does the quoted spread.
The GKN measure corrects for positive autocorrelation in the expected returns, thereby
overcoming the problem that the Roll measure often produces negative spread estimates. In the
framework of George et al. (1991) the logarithm of transaction price at time t can be written as
m is the logarithm of the true value of the asset at time t, π is the proportion of the quoted
spread that is due to order-processing costs,
is is the quoted spread and
it q is an indicator
variable that equals 1 if the transaction at time t is at the ask price and -1 if the transaction is at
the bid price. The true value of asset i consists of the expected return prior to transaction t, the
asymmetric information component, which reflects information revealed by transaction t, and a
white noise term. The logarithm of the bid price after transaction t is
Subtracting the bid price from the transaction price and taking the first difference yields
it it itit it itit
15 See Madhavan (2000) for a more technical survey on the empirical estimation of transaction costs.
Note that equation (3) does not depend on the true value
m and hence any time series properties
that the expected return
may exhibit do not influence
itr . Taking the autocovariance
itr yields the spread measure
i it iti
Using the spread measure
one can infer the proportion of the order processing component
π by a cross-section regression of
According to equation (4) we expect
β = and
Since we do not have bid and ask quotes for the year 1890 to correct for the positive
autocorrelation in expected returns we also employ another measure proposed by George et al.
(1991). For this measure the returns of closing prices, itr , are regressed on the expected return on
the equal weighted market index
E rI− :
r E rI
Then the Roll measure is applied to the residuals of this regression:
i it it
Closely related to the cost of trading is the concept of market liquidity. While the spread itself is
a widely used measure of market liquidity, it cannot reflect quantity reactions to changes in
prices or spreads. Characterizing market liquidity in this manner requires alternative measures,
three of which can be calculated with our historical data: i) the number of trade observations, ii)
the trading volume, and iii) the Amihud illiquidity measure.16 Asset pricing models have found
the Amihud measure particularly useful (e.g. Amihud, 2002, Acharya and Pedersen, 2005, Pastor
and Stambaugh, 2003).
The Amihud stock illiquidity measure can be defined as the average ratio of the daily
absolute return to the absolute dollar volume on that day, i.e.
, where T
defines the averaging period (either monthly or annual). Economically, the illiquidity measure
can be interpreted as the daily price impact caused by the respective order flow.
We test our hypotheses on liquidity, transactions costs, and spread components, using a new
database containing the transaction data for all stocks listed on the New York Stock Exchange
and reported in the New York Times for every trading day (Monday through Saturday) in the
years 1890, 1900, and 1910. We gathered all data reported daily, including closing transaction
prices, closing quoted bid and ask prices (only available in 1900 and 1910), and the number of
shares sold for each stock each day.17
The original New York Times reports contained some errors. For some observations,
errors were easily apparent, and we corrected them. For others, however, it was not as clear-cut.
In these cases we adopted the following procedure: whenever the distance of a particular
observation to the mean of the series exceeded eight times the standard deviation, we treated the
entry as an erroneous datapoint (or at least an extreme outlier observation) and consequently
16 For a more extensive discussion of alternative measures of liquidity see Amihud (2002).
17 The exchange operated every day but Sunday up until 1952, when the Saturday sessions ended. We also collected
the closing price and days’ volume for 1890, but because the NYT did not publish quoted bid and ask prices at that
time, we cannot calculate quoted spreads or conduct the spread decomposition for 1890.
Comparing the data over the three points in time (Table 1), it is clear that the total
number of traded NYSE securities changes only slightly over time, but the number of companies
traded on the NSYE did change considerably from one decade to the next. The number of
companies actually dropped between 1890 and 1900, from 231 to 190, before rebounding
slightly to 200 in 1910.18 While on the face of it, we might expect more shares to enter trading
over time, the introduction of listing requirements in 1895/1896, particularly the obligation to
publish annual reports made formal in 1899, likely depressed numbers. In addition, the crisis of
1893, the drying up of the new issues market from 1893 to 1897, and the beginnings of the
merger wave in 1895, meant the exit of some existing companies and entry of fewer new ones.
Due to an increase in issuance of preferred stocks, however, the number of securities (as opposed
to companies) listed in the New York Times remained quite constant: 326 in 1890, 307 in 1900
and 332 in 1910. Between 1890 and 1900, preferred shares increased as a proportion of total
listings from 23% to 35% and over the following ten years, their proportion relative to common
stocks remained constant. Of these 1910 listings, only 185 (56%) were already traded in 1900,
which also means that about 40 percent of the 1900 listings left the NYSE listings by 1910.
By certain measures, shares traded more actively over time, particularly in the early part
of the period. For example, the average (median) number of trading days rose from 81 (31) per
company in 1890 to 127 (96) in 1900. That number then fell back slightly to 117 (90) in 1910.
The number of companies with at least 90 trading days followed a similar pattern, increasing
from 110 in 1890 to 159 in 1900, and then rose slightly to 166 in 1910. While only 77 firms
traded at least 150 days in 1890, 127 firms did so ten years later (down slightly to 124 firms in
18 These figures are approximate, because the New York Times may have varied its reporting practices over time
and (to a lesser extent) because we may have missed some firms that changed name or for which the New York
Times changed its abbreviation of the name, and we failed to spot it. The abbreviations varied considerably, but we
cleaned the data for this problem as best as possible.
Daily trading volumes follow a different pattern: the median number of shares traded
dropped from 255 in 1890 to 83 in 1900 but then rebounded to 267 in 1910. But since trading
days increased so much between 1890 and 1900, the total trading volume (in $) over the course
of the year still rose. Based on averages, daily trading volume remained fairly constant at around
2,000 shares per day. In 1910, the daily averages covered a wider range, but the number of days
traded increased. But while annual total shares increased significantly for 1910 median dollar
volume fell. Average dollar volume increased by about two-thirds.
The high average volumes relative to medians come from a small number of heavy
traders, such as Reading Railroad, which posted average sales of $14.6 million per day in 1910
(Table 2). The highest 20 trading volumes for each year reached well into the hundreds of
millions annually, with daily averages ranging mostly between a half million and 1.5 million in
1900. The companies with the highest total trading volume change over the three years but three
firms remained among the 20 most traded stocks in 1890, 1900 and 1910 (Union Pacific;
Atchison, Topeka and Santa Fe; and Northern Pacific). One or two highly active shares exceed
all others by a wide margin. American Sugar Refining topped the list in both 1890 and 1900,
with nearly $1.6 billion of shares traded over 287 trading days in the latter year. The Reading
held top place in 1910, with $4.3 billion of shares traded over 293 trading days. The second and
third most traded stocks in 1910, Union Pacific and U.S. Steel, came in slightly behind, at $3.6
and $3.1 billion, respectively, over almost as many trading days. Below that, however, the
annual volume drops off rather quickly, so that the tenth highest volume is about $250 million in
1910, for example. While the enormous railroads clearly dominated the top twenty in dollar
volume, a few of the industrials and utilities—sugar, copper, steel, telephone, and gas—were in
the same league.
V. Overall Market Liquidity and Transactions Costs
We add to this basic picture of market liquidity using quoted bid-ask spreads, a measure of the
effective bid-ask spread (GKN) developed by George et al. (1991), and the Amihud (2002)
illiquidity measure. Given the lack of bid and ask quotes for 1890, quoted spreads and GKN can
only be calculated for 1900 and 1910. Therefore, we also calculate a second measure of the
effective spread, proposed by George et al. (1991), that we denote as GKN2.
A first look at the data presents a somewhat mixed and even contradictory picture. Based
on the GKN2 measure, effective trading costs fell dramatically from a sample-wide average of
3.8% in 1890 to 1.04% in 1900 and then further to 0.82% in 1910. Similarly, market illiquidity
fell markedly over the three points in time. These patterns match well the path of increasing
trading volume for common stocks but contradict the non-monotonic development of trading
volume for preferred stocks. Also, quoted bid-ask spreads increased throughout from 1900 to
1910 for both common and preferred stocks alike. This is particularly true for mean spreads but
also applies to median spreads. The average quoted spreads over the full sample as well as for
most volume categories rose between 1900 and 1910, from about 2.4 percent in 1900 to almost 3
percent in 1910. Median spreads were significantly lower than averages but increased slightly
for the full population, from 1.64 to 1.7 percent (Table 3).19
While improving communications technology could have lowered transactions costs,
other factors seem to outweigh these potential cost reductions. In fact, our finding squares with
Garvy (1944) and Brown et al. (forthcoming) who claim that after 1909 competition from the
Consolidated Exchange declined. This loss of competitive pressure likely allowed brokers to
raise costs. The population of securities clearly changed somewhat over the interval of our
19 Note that Table 3 uses the average spread for each firm, thereby equally weighting all stocks. This method gives
higher means and medians, since they give as much weight to light traders (with higher spreads) as heavy traders
(with lower spreads).
observations, and newer listings carried higher spreads. Notably, however, the phenomenon of
rising quoted spreads still appears among the subset of shares traded in both years (the average of
which increased from 2 percent to 2.6 percent) and among the subset of shares traded on NYSE
and the Consolidated (the average of which increased from 1.3 percent to 1.8 percent).
Preferred shares differ substantially from common stock, particularly in the voting and
dividend rights attached to them. We may expect lower risk and therefore potentially narrower
spreads on preferred shares. According to the GKN2 measure, in all three years preferred shares
have lower effective spreads than common stocks although this difference is small in 1900. It
should be noted, however, that this measure of effective spreads is noisy, and therefore a more
rigorous analysis is deferred to quoted spreads and George et al. (1991)’s other measure of
effective spreads. For 1900 we observe this phenomenon, with preferred shares averaging
quoted spreads of a little more than 1.5 percent, while common stocks averaged nearly 2 percent
(1.96)—a statistically very significant difference.20 In 1910, however, the gap between common
and preferred spreads disappeared almost entirely.
As is often the case in more recent data, effective spreads are lower than quoted spreads
on average in 1900, but the difference is fairly small (2.27 versus 2.42).21 In 1910, the GKN
measure actually exceeds the average quoted spread by a small margin (3.04 versus 2.99). The
estimates of the GKN spread (Table 3), like the quoted spreads, rose significantly between 1900
and 1910—from an average of 2.28 percent in 1900 to an average of 3.04 percent in 1910. As
with the quoted spreads, the distribution of spreads is skewed, so that median effective spreads
are lower, at 1.6 and 1.9, for the same years. Effective spreads also grew on average for the full
20 Based on averages across the entire sample of spreads, rather than using the average within each stock, as reported
in Table 3. Averaging across averages for each stock, the spreads are about 1.8 and 2.8 for preferred and common,
21 We calculate the GKN measure for all companies for which the serial covariance of the returns of the bid prices
could be computed based on at least 4 observations, i.e. where the number of observations of the second order
difference of bid prices is greater or equal to 4. This condition leaves us with 203 firms in 1900 and 182 in 1910.
sample. As with quoted spreads, common stocks trade with higher effective spreads on average
than do preferred shares (2.6 versus 1.8).
While the hypothesized relationship between transactions costs and the liquidity of the
market for a given company’s shares cannot be verified in the inter-temporal evolution of
transactions prices, we check whether it is satisfied in the cross-section within a given year.
Hence, we divide up the sample into thirds based on the average dollar volume of shares traded
each day for the individual shares. Spreads and other characteristics of the stocks differ
substantially depending on the average daily dollar volume traded. Those in the lowest volume
tercile trade with the highest average spreads (both quoted and effective) and also have the
lowest prices and highest return variance. Effective spreads are actually higher than the quoted
bid-ask spreads for the most actively traded stocks A big portion of the increase in spreads
between 1900 and 1910 shows up in the lowest volume tercile, where the GKN measure
increased from an average (median) of 3.4 (2.8) to 4.6 (4.3). Among the most actively traded
stocks, the median GKN measure increases from 0.9 to 1.2 percent. At the same time, the mean
for this tercile rose from 1.3 to 1.8 percent. In other words, in the annual cross-sections,
transaction costs relate monotonically negatively with trading volume for both common and
The estimates for the actively traded stocks are comparable to recent estimates for the
NYSE/AMEX exchanges. For example, Hasbrouck (2006) reports mean trading costs of about 1
percent and a median of .54-.61 depending on the specific measure in use for daily data from
1993-2005. Our estimated effective spreads for 1900 and 1910 are well below Lesmond’s
(2005) estimates for modern emerging markets. The 1890 estimates of the effective spread for
high volume stocks of about 2.2% percent for the mean and 1.5% median are comparable to the
estimates of the Roll measure for China (1991-2000), the Philippines (1987-2000), Portugal
(1988-2000) and Israel (1993-2000). This is another indication for the high degree of trading
efficiency and liquidity in the early and unregulated New York Stock Exchange.
The spread premium for common over preferred shares also shows up most among the
lightest traders, for whom average GKN spreads are nearly twice as high for common stocks as
for preferred stocks in 1900. In fact, the reverse is true among the most active traders—common
stocks trade at similar or even lower average GKN spreads (1.3 versus 1.7 in 1900); though
medians are nearly identical.
In historical comparison the evolution of the Amihud illiquidity measure may be of
particular interest. This measure decreased dramatically from a level of 19.8 in 1890 to 10.1 in
1900 to 5.0 in 1910 for common stocks and 10.8 (1890), 2.2 (1900) to 2.5 (1910) for preferreds.
The corresponding numbers in the high volume tercile of common stocks, however, are
significantly lower: 0.49 in 1890, 0.34 in 1900, and 0.31 in 1910. Surprisingly, those numbers
compare directly with corresponding measures of market illiquidity in the most developed
modern stock markets. Amihud (2002), for example, reports a cross-sectional mean of 0.34 for
NYSE stocks in 1963-1996.22 Thus, the 50 or so highest volume securities at the beginning of
the century traded with illiquidity or transactions costs comparable to the average NYSE trading
costs at the end of the century. Lesmond (2005) reports Amihud measures of 0.39 for China
(1991-2000), 0.43 for the Czech Republic (1993-2000), and 0.46 for Mexico (1988-2000). By
these measures, the most active NYSE stocks at the turn of the century appear slightly less
illiquid than these modern emerging markets.
We can learn even more about the sources of transactions costs and illiquidity by decomposing
the effective spreads into an order processing and asymmetric information components using a
22 Hasbrouck (2006) reports a mean of 0.36 (median of 0.07) on daily cross-sections AMEX/NYSE, 1993-2005.
cross section regression, based on equation (5). We evaluate this equation separately for
common and preferred stocks and also by high and low volume (Table 4). For common stocks,
we find that half of the effective spread comes from order processing costs in 1900, and that
figure falls to about 17 percent in 1910. On the flip side, therefore, the asymmetric information
component rises from 50 percent in 1900 to 83 percent in 1910. Even with the increase in dollar
spreads between 1900 and 1910, the order processing component drops from about 44 cents to
41 cents. For preferred shares, we find an order processing component of 29 percent in 1900,
and that figure jumps to 42 percent ten years later (in dollar terms, 33 to 74 cents). These
patterns fit with the changes in trading volume: increasing for common stocks and decreasing
The spread decomposition also differs depending on trading volume, particularly for
1900. For stocks in the top half of the volume range, the spread is due almost entirely to order-
processing costs. The order-processing component of spreads of firms in the lower half of the
trading volume range is 46 percent in 1900 and 22 percent in 1910. The remainder, of course, is
attributable to asymmetric information. In other words, asymmetric information costs contribute
essentially nothing to the spreads for heavily traded stocks in 1900, while such costs make up
half of trading costs for lighter traders in the same year and 80 percent for both high and low
volume stocks in 1910.
This result contrasts with findings by George et al. (1991) and Stoll (1989) who find that
although the size of the spread varies according to the liquidity of a stock, the composition of the
spread does not. Since trading in stocks should have increasing returns to scale (e.g. due to fixed
costs) we would expect a decrease not only in total spread but also in the order processing cost
component as trading volume increases. However, the results show that asymmetric information
costs essentially disappear as a component of transaction costs for highly traded stocks in 1900
but become much more important in 1910. The fact that the order processing component
increases with volume suggests that spreads contain monopoly rents in 1900.
Also of note, the intercept in the estimation of equation (5) is positive. While
positive and significant in both years, the estimates vary considerably. The generally positive
constant implies that there is negative serial correlation in the adjusted returns even in the
absence of a bid-ask spread. According to Harris (1990) this can be explained by price
discreteness. Since stock prices are expressed as multiples of a minimum tick size there are
rounding errors that increase the negative serial correlation of returns. The minimum tick size on
the NYSE in 1900 and 1910 was $1/8. Figure 1 shows the frequency of quotes on the eight
possible $-fractions. Assuming that the true value of the stock is a continuous variable we would
expect quote prices to be equally distributed among the eight fractions. Figure 1, however, shows
that in both years more than one half of all quotes are integers and around 20 percent end on half
fractions. With more than 70 percent of all quotes being multiples of one half, market makers are
clearly not exploiting the full range of the price grid. This does indeed lead to considerable
rounding errors and a positive intercept in the above regressions.
Moreover, Christie and Schultz (1994) claim that avoiding odd eighth quotes is an
indication for anticompetitive behavior on the part of the market makers. Following this
interpretation we measure anti-competitiveness by the proportion of quotes that are multiples of
one half. Table 5 reports the results of an OLS regression of this measure of anti-competitiveness
on proxies for trading activity (volume and the number of trading days) and risk (variance and a
dummy variable for preferred shares). In addition, we expect a positive relationship between the
proportion of integer or half-integer quotes and the stock price since exploiting the full price grid
is more important for reducing relative rounding errors on stocks with low prices. This effect is
confirmed by our analysis; in both years, the coefficient of log(P) is highly significant and
positive. But even after controlling for this effect, the proxies for trading activity are important;
both volume and the number of trading days relate negatively to price-discreteness. Price
discreteness also increases with risk: Higher return variance significantly increases the mass on
zero and one half price fractions in both years; and preferred shares, which are considered to be
less risky than common shares, have a lower proportion of integer and half-integer quotes (the
difference is significant for 1900). In other words, anti-competitive behavior appears to be more
prevalent for less actively traded and more risky stocks. Another interpretation for this finding is
that market makers do not explicitly quantify their costs but rather use a rule of thumb according
to which they respond to uncertainty by rounding to the next integer or half-dollar price. We
also find, however, that stocks cross-listed on the arch-rival Consolidated Exchange trade with
significantly less price discreteness than those with NYSE-only trading. This finding supports
the interpretation of price discreteness as an indicator of anti-competitive behavior.
VI. Explaining Transactions Costs and Market Liquidity
Spreads and liquidity measures vary considerably among stocks at any point in time, and the
characteristics of securities, and of the trading process for them, may influence these costs.
Thus, we also examine the cross sectional determinants of spreads and of the Amihud measure.
Factors Influencing Spreads (Cross-Sectional Variation)
According to the early view (Demsetz, 1968), market makers simply provide the service for
immediacy. The competitive (realized) bid ask spread compensates for the cost of providing this
service. According to this view the cross sectional variation can be captured by regressions of the
form (Madhavan, 2000):
() ( )
where is denotes security i’s spread.
M denotes market capitalization and proxies firm size,
1 is the inverse of the price as a proxy of the discreteness of price changes. The
underlying riskiness of security i is measured by the volatility of past returns
σ . Trading
activity is measured by volume i V . Studies of modern (i.e. post-WWII) markets reveal that
volume, risk, price and firm size explain most of the variability of bid-ask spreads. Volume tends
to reduce spreads since dealers can turn around inventories more quickly, reducing inventory
risk. Risk typically increases spreads.
Modern market microstructure theory adds to the early view of cost components
based on privileged (or inside) information and dealers’ optimal inventory behavior. So, for a
prominent example, Stoll (2003) delineates between two views—not mutually exclusive—of
transactions costs as a reflection of a market maker’s processing costs and inventory risk (using
real resources) and as compensation for a market maker’s losses to informed traders (not using
real resources). Theoretical models of inventory risk, such as Stoll (1978), find that the
proportional spread is an increasing function of dealer’s risk aversion, the risk (i.e., variance) of
the security being traded, and of the size of the transaction. Theories of information asymmetry
(e.g., Glosten and Milgrom, 1985 and Kyle, 1985) indicate that the spread increases in the
probability of encountering an informed trader as well as in the degree of uncertainty over asset
It is also worthwhile mentioning Dennert’s (1993) work on competing market makers
under adverse selection. While inside traders tend to camouflage their trades in each individual
trade, they can exploit market fragmentation and submit trades almost simultaneously. Market
makers will only realize the activity of insiders, when they find it difficult to sell off their
position in the inter-dealer market. According to this model, volume increases when insiders are
active. This increase correlates positively with the number of market makers in the security.
Hence, according to this theory, insiders will profit more from insider information in actively
traded securities where relatively many market makers provide liquidity. This model predicts
that skewness of volume correlates with insider activity.23 An increase in (positive) skewness of
volume would suggest a reduction of insider trading activity, and hence a reduction in adverse
selection costs, implying a negative correlation between skewness and spreads.
Since inventory holding costs and information affect price dynamics differently, the two
components can be identified if sufficiently rich data are available. Inventory holding costs imply
mean reversion since dealers constantly trade back to their desired inventory position.
Information has permanent effects on security prices since it affects a security’s fundamental
valuation. Glosten and Harris (1988) and Hasbrouck (1999) provide analyses of how the
observed spreads can be decomposed into its components, when data are available on a high
To summarize, the theoretical literature on transactions costs suggest that spreads vary in
cross section, particularly with factors that indicate a stocks liquidity, risk, or information
transparency. For example, the following factors should relate to spreads: in the positive
direction, variance of returns and lumpiness of trading; in the negative direction, stock price,
number of trading days, total trading volume, and proxies for information availability (such as
firm size or age), as well as skewness of volume.
Several empirical studies support these models. In his survey of existing cross-sectional
evidence, Stoll (2003) finds that fundamental values—share volume, return variance, price,
number of trades and market value—nearly always relate very significantly to spreads in modern
data. Trading volume is particularly informative: in both theory and practice, heavily traded
stocks trade with lower costs. These stocks benefit from economies of scale in order processing,
23 This prediction has largely been unexplored in empirical work so far. Madhavan, for example, does not refer at
all to Dennert (1993). On the other hand, Dennert does not emphasize the empirical implications of his model.
significantly lower inventory holding costs and illiquidity risk to market makers, and possibly
also from greater transparency of information regarding the underlying company and therefore
lower asymmetric information costs. To the extent that they tend to also have more outside
shareholders, the likelihood of encountering an insider is also reduced. If specialists retain
market power even in high-volume stocks, however, we could anticipate little or no decline in
the order processing component of the spread.
Table 6 sums up the testable implications of the various theories of factors
influencing bid-ask spreads. Certain empirical variables proxy for more than one theoretical
relationship. For example, trading volume and number of trading days provide an indication of
the liquidity of a security’s market and may also proxy for information availability; they may
also relate inversely to the probability of a market maker encountering an insider trade. For
some variables, such as trade size, individual transaction data are unavailable for the period of
Hypotheses on historical average spreads and components
We set out several hypotheses based on the historical analysis. First, at the narrow level of the
individual specialist, there seem to have been opportunities to set transactions costs above
competitive levels due to the market power exercised, at least in some shares. At the broader
level, the continuous market mechanism could have reinforced this tendency by splitting up
orders into smaller lots (as opposed to aggregating the day’s trades into larger ones) and thereby
keeping order processing costs high. As communications technologies had advanced relatively
far by 1900, order processing costs may have declined from their 19th century levels. Moreover,
the NYSE maintained relatively tight listing standards, which should have mitigated asymmetric
information costs. Still, we expect that the cost increasing factors dominate, so that we should
find that average total transactions costs (realized spreads) for the full population of stocks in
1900 and 1910 exceed those calculated for the post-WWII period, particularly the last few
decades of the 20th century. We hypothesize further that these relatively high costs stem both
from a high order processing component and a somewhat higher adverse selection component.
In terms of the cross sectional variation in spreads, we expect that the theoretical models
apply in the earlier stages of market development as they have in recent years. Thus, our
hypotheses remain essentially the same as those posited in the literature.
In this final section, we look at the cross-sectional determinants of bid-ask spreads and of the
Amihud measure . While there is no single structural model of trading costs, the various
theoretical models of spreads lead to a number of testable predictions, many of which are
summarized in Table 6. We use the following variables to explain average percentage spreads:
average daily dollar volume (V), average price (P), variance of stock returns (Var) and the
number of trading days (Days). This specification largely parallels Stoll (2000) and other studies
surveyed by Madhavan (2000) and Stoll (2003), notably Demsetz (1968). Additionally, we
include the skewness of daily trading volume, in order to attempt to proxy (inversely) for the
lumpiness of trading, an indicator that relates to the frequency of large trades and therefore the
likelihood of the market maker dealing with an informed insider.
For each year (1900 and 1910) and each stock type, common versus preferred, we run
separate regressions based on monthly averages of our data.24 Aggregating our data to a monthly
frequency helps to eliminate the noise of daily data and still leaves a rich enough structure to
allow for dynamic interactions between spread and liquidity measures and also the computation
of return variance and volume skewness. Since transactions prices, volume and liquidity are
24 Recall that the data sources available do not quote bid and ask prices for 1890, so we cannot estimate the quoted
spread model for that year.
theoretically jointly determined, we only consider predetermined variables as regressors in order
to reduce the impact of simultaneity on our parameter estimates.25 Moreover, we use median
regression to deal with outlier problems that are prevalent in our data set (see Figure 2).
As hypothesized, illiquidity relates positively with spreads, while trading volume (both
measures) and the price level relate negatively with spreads (Table 7). These cross-sectional
relationships hold for both common and preferred stocks in both years. Interestingly, the
elasticity of spreads with respect to volume is the same for common and preferred stocks, about
2% in 1900 declining to 1.1% in 1910. The elasticity of spreads with respect to the price level,
however, is twice for preferred relative to common stocks in both years. Less consistently, but in
line with theory, risk has a positive impact on spreads. Skewness of trading volume on the other
hand does not seem to systematically drive spreads.
Interestingly, we find a strong negative impact of cross-listing on the Consolidated
Exchange, particularly for common stocks.26 The cross-listing dummy measures the extent of the
competitive pressure that the Consolidated Exchange exerted on New York Stock Exchange
quoted spreads. This finding significantly strengthens the result of Brown et al. (forthcoming),
who identify the competitive effect in the time domain around the emergence and the closure of
the Consolidated Exchange. The evidence is less clear for preferred stocks; however, only few of
those are dually traded. We also find that rail stocks tend to have tighter spreads. Rail stocks are
generally traded on both exchanges in larger volumes. Hence, we may be picking up an
additional effect of the competitive cross market trading in high-volume shares.
The determinants of illiquidity essentially are the same as those of quoted spreads.
However, our results confirm that the Amihud measure aggregates those factors differently
25 We tested various cross-sectional regressions based on annual data and found ample evidence of simultaneity
problems and parameter instability.
26 Strictly speaking “cross-listing” really means trading on both exchanges, since the Consolidated Exchange did
not have a formal listing procedure. In principle, any security could be traded there. In 1900, preferred shares have
a negative cross-listing effect, but in 1910, the effect is reversed.
relative to the quoted spread. Still, volume has a negative impact on illiquidity, while the spreads
and the price level consistently exert a positive impact on illiquidity. Rail companies have lower
illiquidity only for their common stocks and only in 1900; otherwise the rail dummy is irrelevant
for explaining illiquidity. Likewise the number of trading days only affects illiquidity of common
stocks in 1900. Skewness of volume has no discernible impact.
The cross-listing dummy does not exert a systematic influence on liquidity. Statistically,
it has a strong positive impact only in 1900 and a negative impact on preferred stocks in 1910.
From a theoretical perspective two forces of dual trading are interacting in this case, leaving the
aggregate effect unclear: dual trading tends to reduce liquidity in each market, while competition
and lower trading costs tends to boost liquidity. In general, the net effect cannot be predicted
without further information about market characteristics.
Overall, the cross-sectional results on the illiquidity measure accord well with those on
the quoted spread. However, explanatory power is significantly higher for the spread regressions.
Moreover, these results on a key historical market fall very much in line with similar cross-
sectional analyses of established (e.g. Stoll, 2003) or emerging markets recently (Lesmond,
2005).27 In this sense, we argue that the behavior of traders, and therefore the drivers of price
discovery and liquidity, are already discernable in this unregulated regime and are quite
comparable to those in tightly regulated modern markets.
This paper contributes in several ways to the newly emerging line of research into the
microstructure and performance of securities markets in the unregulated era. We provide the
27 Our cross-sectional parameter estimates correspond reasonably well with Lesmond’s (2005) estimates for modern
emerging markets. While he can control for legal origin in his data set, in our historical analysis we can verify a
strong negative influence of the quoted bid-ask spread.
first comprehensive daily measures of market illiquidity, quoted bid-ask spreads, and effective
transactions costs for the pre-WWI NYSE; the last of which we decompose into order processing
and asymmetric information components. We also analyze changes in these measures over time
as well as the cross-sectional determinants at each point in time.
Most interestingly, we find that trading costs and measures of illiquidity were roughly
comparable with modern-day rates for the most heavily traded securities. This finding
demonstrates that even prior to the introduction of regulatory oversight, early securities markets
did perform remarkably well. Moreover, the decomposition of the quoted bid-ask spread into
order-processing and an adverse selection components suggests that adverse selection is not a
significant component of trading costs in the earlier periods. While order processing costs did not
vary much from 1900 to 1910 in absolute terms, their relative role declined. Moreover, we find
evidence that quoted spreads did react to competitive pressure from rival markets, such as the
Consolidated Exchange. NYSE securities that did not trade on the Consolidated exhibit a
substantially larger probability of price-clustering, an indicator of non-competitive conduct. It is
likely that the competitive reaction for dually-traded securities is largely driven by the order-
processing component. On the other hand, fragmented trading should increase the adverse
selection component. The analysis of the precise nature of competition between those early
exchanges requires even more detailed data and, therefore, is left for future research.
Of methodological importance, we find different patterns in the various measures of
trading costs and market liquidity. While we find that mean measures of market illiquidity,
trading volume, and effective spreads all move together over time, quoted bid-ask spreads do
not. Thus, the first three measures may provide a more accurate picture of market functioning at
the aggregate level than do the quoted spreads. Cross-sectional analysis of the determinants of
quoted spreads, however, indicate that this measures does help differentiate among individual
stocks at a given point in time, reacting as expected to measures of trading volume, price, and
risk. The measure also relates closely to illiquidity in cross section, and the determinants of the
quoted spread are more robust.
Our analysis of the performance of the early New York securities markets suggests little
cause for the regulatory intervention at least by that time. Based on our decomposition of quoted
prices we find some evidence that informed trading was becoming a more serious component of
trading costs from 1900 to 1910. Further research (and much more data collection) is needed to
examine the pre-regulatory era completely. Our work suggests that market microstructure
analysis based on historical cross-sectional trading data may prove useful to a scholarly analysis
of the regulatory process that ultimately generated the various regulatory instruments such as
insider trading restrictions and conduct regulation to ensure competitive pricing. Of particular
use would be analysis of specific crises that prompted arguments leading to the foundation of the
Acharya, Viral, and Lasse Heje Pedersen. 2005. “Asset Pricing with Liquidity Risk.”
Journal of Financial Economics, 77: 375-410.
Allen, Franklin Lubomir Litov, and Jianping Mei. 2006. “Large Investors, Price
Manipulation, and Limits to Arbitrage: An Anatomy of Market Corners.” Review of
Finance, 10(4): 645-693.
Amihud, Yakov. 2002. “Illiquidity and Stock Returns: Cross-Section and Time-Series Effect.”
Journal of Financial Markets, 5: 31-56.
Archambault, Jeffrey J., and Marie Archambault. 2005. “The Effect of Regulation on
Statement Disclosures in the 1915 Moody’s Manuals.” Accounting Historians Journal,
Baskin, Jonathan B. 1988. “The Development of Corporate Financial Markets in Britain and
the United States, 1600-1914: Overcoming Asymmetric Information.” The Business
History Review, 62: 199-237.
Brown Jr., William O., J. Harold Mulherin and Marc D. Weidenmier. Forthcoming.
“Competing with the NYSE.” Quarterly Journal of Economics.
Calomiris, Charles W., and Carlos D. Ramirez. 1996. “Financing the American Corporation:
The Changing Menu of Financial Relationships.” Columbia Business School and George
Mason University Working Paper.
Carosso, Vincent P. 1973. “The Wall Street Money Trust from Pujo through Medina.” The
Business History Review, 47(4): 421-437.
Christie, William G., and Paul H. Schultz. 1994. “Why Do NASDAQ Market Makers Avoid
Odd-Eighth Quotes?” Journal of Finance, 49: 1813-1840.
Davis, Lance E., and Robert J. Cull. 1994. International Capital Markets and American
Economic Growth, 1820-1914. New York: Cambridge University Press.
Davis, Lance E., and Larry Neal. 1998. “Micro Rules and Macro Outcomes: The Impact of
Micro Structure on the Efficiency of Security Exchanges, London, New York, and Paris,
1800-1914.” The American Economic Review, 88(2), Papers and Proceedings of the
Hundred and Tenth Annual Meeting of the American Economic Association: 40-45.
Davis, Lance E., Larry Neal, and Eugene White. 2007. “The Highest Price Ever: The Great
NYSE Seat Sale of 1928–1929 and Capacity Constraints.” The Journal of Economic
History, 67: 705-739.
Demsetz, Harold. 1968. “The Cost of Transacting.” Quarterly Journal of Economics, 82: 33-53.
Dennert, Jürgen. 1993. “Price Competition Between Market Makers.” Review of Economic
Studies, 60: 735-751.
Garbade, Kenneth D., and William L. Silber. 1979. “Structural Organization of Secondary
Markets.” Journal of Finance, 34: 577-93.
Garvey, George. 1944. “Rivals and Interlopers in the History of the New York Security
Market.” Journal of Political Economy, 52(2): 128-143.
Gehrig, Thomas, and Caroline Fohlin. 2006. “Trading Costs in Early Securities Markets: The
Case of the Berlin Stock Exchange 1880-1910.“ Review of Finance, 10: 587-612.
Geisst, Charles R. 2004. Wall Street: A History from its Beginnings to the Fall of Enron. New
York: Oxford University Press.
George, Thomas J., Gautam Kaul, and M. Nimalendran. 1991. “Estimation of Bid-Asks
Spreads and its Components: A New Approach.” Review of Financial Studies, 4(4): 623-
Glosten, Lawrence R., and Lawrence Harris. 1988. “Estimating the Components of the Bid-
Ask Spread.” Journal of Financial Economics, 14: 21-142.
Glosten, Lawrence R., and Paul Milgrom. 1995. “Bid, Ask and Transaction Prices in a
Specialist Market with Heterogeneously Informed Traders.” Journal of Financial
Economics, 14: 71-100.
Hasbrouck, Joel. 1999. “Security Bid-Ask Dynamics with Discreteness and Clustering: Simple
Strategies.” Journal of Financial Markets, 2: 1-18.
Hasbrouck, Joel. 2006. “Trading Costs and Returns for US Equities: Estimating Effective Costs
from Daily Data.” mimeo, NYU.
Harris, Lawrence. 1990. “Estimation of Stock Price Variances and Serial Covariances from
Discrete Observations.” Journal of Financial and Quantitative Analysis, 25(3): 291-306.
Hawkins, David F. 1963. “The Development of Modern Financial Reporting Practices among
American Manufacturing Corporations.” The Business History Review 37(3): 135-168.
Hickson, Charles R., and John D. Turner. 2005. “The Genesis of Corporate Governance:
Nineteenth-Century Irish Joint-Stock Banks.” Business History, 47: 174-189.
Jones, Charles M. 2002. “A Century of Stock Market Liquidity and Trading Costs.”
Unpublished Working Paper, Columbia University.
Kregel, Jan A. 1995. “Neoclassical Price Theory, Institutions, and the Evolution of Securities
Market Organisation.” The Economic Journal, 105(429): 459-470.
Kyle, Albert. 1985. “Continuous Auctions and Insider Trading.” Econometrica, 53: 1315-1335.
Lesmond, David A. 2005. “Liquidity of Emerging Markets.” Journal of Financial Economics,
Madhavan, Ananth. 2000. “Market Microstructre: A Survey.” Journal of Financial Markets, 3:
Michie, Ranald C. 1986. “The London and New York Stock Exchanges, 1850-1914.” The
Journal of Economic History, 46: 171-187.
Michie, Ranald C. 1987. The London and New York Stock Exchanges, 1850-1914. Boston:
Allen and Unwin.
Mulherin, J. Harold, Jeffry M. Netter, and James A. Overdahl. 1991. “Prices are Property:
The Organization of Financial Exchanges from a Trading Cost Perspective.” Journal of
Law and Economics, 34: 591-644.
Navin, Thomas R., and Marian V. Sears. 1955. “The Rise of a Market for Industrial
Securities, 1887-1902.” Business History Review, 105: 106-112.
New York Stock Exchange Online.
timeline_chronology_index.html. Accessed 4/27/2005.
Pastor, Lubos, and Robert F. Stambaugh. 2003. "Liquidity Risk and Expected Stock Returns."
Journal of Political Economy, 111(3): 642-85.
Roll, Richard. 1984. “A Simple Implicit Measure of the Effective Bid-Ask Spread in an
Efficient Market.” Journal of Finance, 39(4): 1127-1139.
Sobel, Robert. 1965. The Big Board: A History of the New York Stock Market. New York: The
Stoll, Hans R. 1978. “The Supply of Dealer Services in Securities Markets.” Journal of Finance,
Stoll, Hans R. 1989. “Inferring the Components of the Bid-Ask Spread: Theory and Empirical
Tests.” Journal of Finance, 44(1): 115-134.
Stoll, Hans R. 2000. “Friction.” Journal of Finance, 55: 1479-1514.
Stoll, Hans R. 2003. “Market Microstructure.” in Handbook of the Economics of Finance, edited
by G.M. Constantinides, M. Harris, and R. Stulz, Elsevier Science B.V.
Vitols, Sigurt. 2001. The Origins of Bank-Based and Market-Based Financial Systems:
Germany, Japan and the United States, Discussion Paper.
Table 1: Listings and Trading Activity on the New York Stock Exchange, 1890-1910
Number of companies in NYT
Number of issues by share type
Number of trading days for all
Number of companies with at
least N trading days
Number of shares traded daily
Volume traded (dollars) annually Mean
Notes: In the first row, all issues for one company count as one company.
Table 2: The 20 Most Traded Stocks at the NYSE, 1890 - 1910
Panel A: 1890
Sugar Ref. Co
Del., Lack & Western
Chicago, Mil. & St. Paul
Chicago, Rock Island & Pacific
Louis. & Nashville
Phil. & Reading
Atchison, Topeka and Santa Fe
Chicago, Bur. & Quincy
Northern Pacific pf.
Chicago Gas Co.
C., C., C. & St. L.
Chicago & Northwestern
New York & New England
Oregon & Trans.
Western Union Telegraph
Chicago Gas Trust
Rich & W. P.e
Panel B: 1900
American Sugar Ref. Co.
Brooklyn Rapid Tran.
Pennsylvania R. R.
Chicago, Mil. & St. Paul
Atchison, Topeka and Santa Fe pf.
Chicago, Bur. & Quincy
People's Gas, Chicago
Reading 1st pf.
Met. Street Railway
Baltimore & Ohio
American Steel & Wire
New York Central
Chicago, Rock Island & Pacific
Volume traded ($)
No of obs. Avg. volume ($)
Volume traded ($)
No of obs. Avg. volume ($)
Total volume ($)
Total volume ($)
Panel C: 1910
U. S. Steel
Pennsylvania R. R.
American Smelt. & Ref
Chi., Mil. & St. Paul
Atch., Top. & S. F.
N. Y. Central
Chesapeake & Ohio
Rock Island Co
Brooklyn Rapid Tran
Great Northern pf.
American Tel. & Tel.
Volume traded ($)
Avg. volumes ($)
No of obs. Total volume ($)
Table 3 (a): Effective Spreads and Liquidity in 1890
999 9 9
10 10 1010
4343 43 43 43
30 3030 30
Table 3 (b): Quoted and Effective Spreads and Liquidity in 1900
131365 86 31.43
Avg. St. dev. of
7274 74 74
1717 17 17
3739 39 39 3939
1818 1818 18 18
48 5151 51 51 51 51
28 292929 29 29
49 494949 49 49
Table 3 (c): Quoted and Effective Spreads and Liquidity in 1910
volume ($) ing days price
68 68 68
27 27 27
13 13 13
# of trad- Avg. St. dev. of
6868 68 68
28 28 28 28
27 27 27 27
13 13 13 13
32 32 32 3232 32
3434 34 34 3434
48 48 4848 48 48
Table 4: Spread Decomposition Regression and the Implied Components
The dependent variable is the annually estimated spread. The independent variable is the average quoted spread. t-statistics are given in
Table 5: Price Discreteness Regressions
Dependent variable is the proportion of quotes that are multiples of one half.
Variable Coefficient Std. Error*
R-squared 0.7902 Observations:
Variable Coefficient Std. Error*
R-squared 0.7812 Observations:
*White Heteroskedasticity-Consistent Standard Errors & Covariance
Factors increasing spread Available proxy variable Expected relationship
Fundamental risk of security
Variance of returns
Daily or monthly volume/days
Skewness of volume
Transaction size (inventory
Illiquidity of market
Probability of informed trade
Trading volume, stock price
Number of trading days, total
Transaction size (lumpiness)
Skewness of volume
Firm size, age, information
availability, preferred shares
Negative Lack of fundamental
Market power of specialists Positive
Table 7: Determinants of Average Spreads and the Amihud Liquidity Measure
This table presents the results of a least absolute deviations (LAD) estimation. All continuous variables are monthly averages.
Dummy(Rail) equals 1 if the issuing firm was a rail company and zero otherwise. Dummy(Consolidated) equals 1 if more than 100
shares traded on the Consolidated on at least one of 12 randomly selected days and zero otherwise. The estimation corrects for
heteroskedasticity in the error structure using the Huber sandwich standard errors and covariances. t-statistics are given in parentheses.
Spread 1900 Spread 1910
Indep. Variables Common Preferred Common Preferred
0.099 0.101 0.108 0.122
(25.71)*** (15.65)*** (14.96)*** (13.52)*** (1.06)
-0.002 -0.002 -0.001 -0.002
(-8.83)*** (-5.65)*** (-3.46)*** (-2.45)**
-0.004 -0.008 -0.007 -0.011
(-9.68)*** (-8.88)*** (-7.67)*** (-6.56)*** (-0.939)
2.020 -0.139 0.520 3.985
(2.20)** (-0.87) (10.80)*** (2.27)**
30.030 -0.058 -2.381 -320.562
(0.86) (-0.12) (-15.55)*** (-3.65)*** (-0.14)
-0.016 -0.012 -0.019 -0.014
(-15.07)*** (-7.73)*** (-8.86)*** (-9.00)*** (2.03)**
-0.001 -0.002 0.001 -0.001
(-2.93)*** (-3.05)*** (1.55) (-1.64)
0.0004 0.0004 0.0011 0.0008
(5.74)*** (3.13)*** (8.10)*** (16.79)***
-0.001 0.000 0.000 0.000
(-3.31)*** (-0.65) (-0.35) (0.16)
Dummy(Consolidated) -0.0017 -0.002 -0.002 0.002
(-3.02)*** (-2.73)*** (-3.38)*** (1.63)
No. of Observations 1215 725 1298 707
Pseudo R-squared 0.444 0.365 0.401 0.327
Adjusted R-squared 0.439 0.357 0.397 0.318
Figure 1: Frequency of price fractions
0 0.1250.25 0.375 0.5 0.625 0.750.825
Figure 2: Boxplot graphs of the cross-section of monthly average spreads. The box contains the middle
50% of the data. The vertical line inside the box is the median and the black dot is the average. The
shaded area gives the 95% confidence interval for the median. The whiskers show the highest or lowest
value unless there are outliers in which case they extend up or down to 1.5 times the height of the box.
Diamonds and stars denote near and far outliers, respectively. The graphs are cut off at 0.15.
(a): Monthly average spreads of common stocks in 1900:
(b): Monthly average spreads of preferred stocks in 1900:
(c): Monthly average spreads of common stocks in 1910: Download full-text
(d): Monthly average spreads of preferred stocks in 1910: