Transforming the NYSE:
Crisis and Expansion during the Great War
and its Aftermath
Caroline Fohlin
Johns Hopkins University
and
Emory University
Plan of the talk
•
•
•
•
•
•
Motivation
Historical Background
Measuring Market Quality
Data
Empirical Results
Further Work
Motivation
• Transformative period
•
•
•
•
WWI crisis
High inflation
Post-war recession
Rapid technological
change
• Global capital market
upheaval
• Liquidity drain
• War bond issues
• Post-war debt crisis
Motivation
• Crisis and expansion: impact on the NYSE?
–
–
–
–
–
War itself
Global war finance
Rapid increase in listings
New companies/technologies
Post-war recession
• All challenges to market liquidity and stability
Motivation
• Why study market microstructure?
- Bottom-up approach to understanding how well markets function
• Structure of trading and execution institutions
• Government regulation
•
Industry structure (competition)
- Analysis of microstructure can provide some answers about
• Market efficiency
• Asset pricing (Amihud, JFM, 2002)
• Liquidity risk is an independent risk factor (Acharya, Pedersen, JFE, 2005)
• Competition among stock markets/market makers
• Competition between stock exchanges (Gehrig, Stahl, Vives, CEPR, 1995)
• Market power of market participants
• Cross-listing of firms (Domowitz et al. JFE, 2002)
• Behavior of investors
Motivation
•Why study World War I era markets?
• Unique period of development
• Onset of war drained liquidity from global financial markets
• Changing mix of securities
Historical Background
• NYSE founded in 1792
• Rapid development over late 19th
and early 20th centuries
•
•
Dominated by rails until end of 19th
century
Gradual shift to industrials in early
20th century
Historical Background
• NYSE microstructure
•
•
•
Originally a call market
Moved to continuous market in
1871
Emergence of specialists who
bear risks
•
Also gain market power
Historical Background
• NYSE tried to exert market
power
• Limited listings
• Limited memberships
• Fixed minimum commissions
• Spurred local competitors
• NYSE fought to suppress
• Curb
• Consolidated (until mid-1920s)
– NYSE seat cost almost $100k
in 1905; Consolidated seat
cost $500
Market quality
Number of securities traded
Trading volume
Quoted bid-ask spread: cost of a hypothetical round trip
Relative effective spreads
Roll measure (1984): effective bid-ask spread
GKN-measures (1991): correction in case of positively correlated
transaction returns
• Amihud stock illiquidity measure (2002): daily price impact caused by the
respective order flow
• Volatility
•
•
•
•
•
•
Data
• Gathered data for every equity security traded on the NYSE
– used NYT via Proquest
• Weekly frequency (Fridays) 1911 – 1925
– manually double entered, compared, and randomly checked
• All available data
–
–
–
–
–
–
–
name
type of security (preferred, rights, trust certificates),
ex-dividend days, ex-rights days,
volume traded (number of shares, aka “sales”)
first and last, high and low transaction prices
quoted bid and ask prices prevailing at the close of the market
any other details printed in the New York Times stock tables
Trading Activity: Overall
Trend
• Big increase in the number
of distinct securities traded
on a daily basis
•
Rose from 100 to 150 in
1911 to well over 400 in
1920
• Declining proportion of
preferred stock prior to
WWI
•
•
•
Heavy use up to war
Common stocks took over,
especially after war
Topic for further
investigation
Source: Caroline Fohlin (2014)
Trading Activity:
Variability
•
–
E.g., 332 distinct stocks
traded 6/30/1922 but 416
different stocks traded a
week later, on 7/7/1922
Trading volume often ran
50-60 percent higher or
lower than the weekly
average volume
Rise and fall during war
–
Rise over 1st 2+ years;
decline during US
involvement (April ’17 to
November ‘18)
– vast volumes of Liberty
Bonds, price controls,
stock exchange controls,
tax hikes, physical
dislocation of personnel
•
400
Weekly variation
–
•
Number of Distinct Common Stocks
Traded on the NYSE Fridays, 19111922
Big increase after WWI
350
y = 1E-05x2 - 0.0891x + 230.6
R² = 0.8977
300
250
200
150
100
50
Source: Caroline Fohlin (2014)
0
01/06/1911 01/06/1913 01/06/1915 01/06/1917 01/06/1919 01/06/1921
Trading Activity: Cycles
•
Rising volume (number) of
shares traded
•
•
380,000 shares avg. in the
1911-14 period to nearly
870,000 shares avg. in the
1919-22 period
Cyclical variation
–
At least three cycles before,
during, and after WWI
Source: Caroline Fohlin (2014)
Liquidity
•
RES proxies for liquidity
-
•
•
proportional deviation of
prices from mid-quote
High RES  illiquidity
Markets respond to
uncertainty with higher
spreads
Weekly variation up to 25% of
average RES
Source: Caroline Fohlin (2014)
Liquidity
•
•
Quoted spreads also proxy for
liquidity, but less accurate
measure of true transaction
costs
Similar patterns as RES over
the period
Source: Caroline Fohlin (2014)
Volatility
•
•
•
Quasi volatility: (highlow)/last
Obvious break at WWI
Continued rise after WWI
Source: Caroline Fohlin (2014)
Table 1. Subperiod averages
Period
Sales
Total dollar
Median RES Median
Median
(number volume
(common) spread
quasiof shares)
(common, %) volatility (%)
Mean 382,026
1911-14
37,629,136
0.60
0.90
0.62
Median 308,724
30,100,000
0.57
0.87
0.57
Std dev. 228,252
22,865,900
0.15
0.23
0.31
Mean 868,244
1919-22
54,531,319
1.13
1.35
1.53
Median 812,727
50,250,000
1.14
1.43
1.34
Std dev. 359,371
27,897,184
0.23
0.35
0.52
Note: RES means relative effective spread. See text for definitions.
Source: Caroline Fohlin (2014)
Why was the NYSE becoming less liquid?
• Microstructure theory offers some hypotheses
1. Growing valuation uncertainty due to war?
2. Rising asymmetric information problems?
3. Changing population (riskier) of the market?
• Or increasingly binding capacity constraints?
• Loss of competition? Increased fragmentation?
Why was the NYSE becoming less liquid?
• Main culprit is growing valuation uncertainty
–
–
–
–
Higher volatility drives up spreads (aggregate and cross-sectional)
Volatility (in cross section) has growing impact on spreads over time
Spreads mostly fall when prices or median volume rise
Controlling for theorized factors leaves no more upward trend in RES
• Aggregate and individual trading volume not causing capacity
problems
– In post-WWI period, RES rises with both share prices and median
trading volume and falls with median volatility
– But individual (stock-level) spreads fall with own-stock volume
Liquidity Differences across Periods
Relative effective
spread
1 vs 2
1 vs 3
2 vs 3
3 old vs 1
3 old vs 2
3 old vs 3 new
Mean A
Mean B
STD A
STD B
Pvalue <
Pvalue =
Pvalue >
Num Obs A Num Obs B
1.19
1.19
2.03
1.95
1.95
1.95
2.03
2.16
2.16
1.19
2.03
2.55
2.51
2.51
8.48
4.78
4.78
4.78
8.48
4.61
4.61
2.51
8.48
4.27
0.00
0.00
0.01
1.00
0.06
0.00
0.00
0.00
0.02
0.00
0.12
0.00
1.00
1.00
0.99
0.00
0.94
1.00
18,037
18,037
27,032
59,415
59,415
59,415
27,032
92,029
92,029
18,037
27,032
32,614
0.021
0.021
1.019
0.102
0.102
0.102
1.019
0.119
0.119
0.021
1.019
0.167
0.086
0.086
4.559
0.318
0.318
0.318
4.559
0.300
0.300
0.086
4.559
0.241
0.072
0.000
0.906
0.995
0.090
0.066
0.145
0.000
0.188
0.009
0.180
0.132
0.928
1.000
0.094
0.005
0.910
0.934
30.000
30.000
46.000
142.000
142.000
142.000
46.000
194.000
194.000
30.000
46.000
52.000
2.11
2.11
3.92
1.93
1.93
1.93
3.92
2.36
2.36
2.11
3.92
3.14
20.26
20.26
142.78
5.62
5.62
5.62
142.78
11.89
11.89
20.26
142.78
18.46
0.02
0.06
0.96
0.12
0.01
0.00
0.04
0.11
0.07
0.24
0.02
0.00
0.98
0.94
0.04
0.88
0.99
1.00
18,018
18,018
26,969
58,873
58,873
58,873
26,969
91,072
91,072
18,018
26,969
32,199
Amihud
1 vs 2
1 vs 3
2 vs 3
3 old vs 1
3 old vs 2
3 old vs 3 new
qvol
1 vs 2
1 vs 3
2 vs 3
3 old vs 1
3 old vs 2
3 old vs 3 new
Table 2a. Factors Associated with Median RES at the NYSE on Fridays, 1911-1922
Period 1
1911-1914 (July)
Period 2
1915-1918
Period 3
1919-1922
Median last
price
-0.0043***
-0.00325
-0.0057***
-0.00808
-0.00925*** -0.008***
(2.10e-08) 0.00
1.256*** 1.601***
-0.0013
(3.30e-06)
Median sales
(number of
shares)
-0.0003***
(3.88e-05)
-0.0003***
-0.00612
-0.000104** -0.000157***
-0.0146
(3.47e-06)
0.00589
-0.186
0.046***
(9.59e-10)
Total dollar
volume
2.61E-07
-0.738
-1.68E-07
-0.885
-2.06E-07
-0.72
3.12E-07
-0.48
-0.0001
-0.188
-0.00029**
-0.0293
Median quasi- 0.148***
volatility
-0.00839
0.173**
-0.0443
0.166***
(2.27e-10)
0.182***
0
-0.407
-0.864
-22.07***
(9.91e-08)
Trend
0.00047*
-0.0842
0.0004
-0.207
0.000177
-0.457
-2.04E-05
-0.903
-0.500** -0.242***
-0.0154
(1.17e-05)
Constant
0.844***
0
0.919***
(1.25e-09)
1.128***
0
1.080***
0
26.19
-0.416
-12.88
-0.497
Observations
R-squared
Model type
182
0.226
p-w
183
192
0.463
p-w
193
185
0.131
p-w
186
median
median
median
Note: P-values beneath coefficients. “p-w” indicates Prais-Winston regression (implemented in Stata),
while “median” indicates median regression. The latter procedure does not produce r-squared statistics.
Source: Caroline Fohlin (2014)
Table 2b. Factors Associated with Median Quoted Spreads at the NYSE on Fridays,
1911-1922
Period 2
Period 3
1915-1918
1919-1922
Period 1
1911-1914 (July)
Median
last price
-0.00988***
(2.17e-07)
-0.0120***
(1.12e-06)
-0.011***
(1.21e-08)
-0.0097***
0
-0.029***
0
-0.024***
0
Median
sales
(number
of shares)
-0.00045***
(2.94e-07)
-0.00055***
(8.93e-06)
-0.00026***
(3.90e-08)
-0.00025***
0
-0.00015**
-0.0124
-0.00024***
(6.69e-05)
Total
dollar
volume
1.66e-06*
-0.094
2.17E-06
-0.107
-1.29E-07
-0.839
-5.32E-07
-0.191
1.73e-06*
-0.0964
8.89E-07
-0.386
Median
quasivolatility
0.242***
-0.000699
0.237**
-0.0157
0.275***
0
0.281***
0
0.160***
(2.88e-06)
0.198***
(1.35e-08)
Trend
0.000909**
-0.0107
0.00085**
-0.0285
0.00038
-0.189
0.0004**
-0.0116
-0.0029***
(1.87e-06)
-0.0024***
(2.07e-07)
Constant
1.385***
0
1.515***
0
1.496***
0
1.418***
0
2.671***
0
2.462***
0
Observati
ons
R-squared
Model
type
182
183
191
192
185
186
0.396
p-w
qreg
0.579
p-w
qreg
0.551
p-w
qreg
Source: Caroline Fohlin (2014)
Table 3. Factors Associated with Individual Relative Effective Spreads and with
Percentage Quoted Spreads at the NYSE on Fridays, 1911-1922
Period 1
-0.005
0.000
RES
Period 2
-0.004
0.000
Period 3
-0.008
0.000
Quoted spread (%)
Period 1
Period 2
Period 3
-0.008
-0.005
-0.010
0.000
0.000
0.000
Sales
-0.0001
0.000
-0.0002
0.000
-0.0003
0.000
-0.0001
0.000
-0.0003
0.000
-0.0006
0.000
Quasivolatility
0.002
0.289
0.040
0.000
0.056
0.000
-0.010
0.012
0.053
0.000
0.066
0.000
Year
0.009
0.074
-0.002
0.869
-0.011
0.075
0.052
0.000
-0.002
0.819
-0.025
0.000
Constant
1.007
0.000
0.992
0.000
1.570
0.000
1.457
0.000
1.373
0.000
2.240
0.000
Observations
17,233
15,470
44,579
17,048
15,314
44,196
Last price
Note: Periods are 1911-1914 (July); 1915-1918; 1919-1922. Equations are estimated using quantile
regression.
Much more to come…
• Working to carefully track
individual securities through
time
 Panel studies of changing
market quality
 Roll refinements (e.g. GKN
and LOT measures)
 Decompose spreads to
measure adverse selection
costs
 Liquidity factor in asset
pricing
Source: Caroline Fohlin (2014)
Much more to come…
• Finish incorporating in
1923-25
 Analyze the continuing
(even greater)
expansion
 Link up to CRSP in 1926
and study lead-up to GD
and then entire 20th
century
Source: Caroline Fohlin (2014)
Key points to take away
• WWI set the stage for transformation of the NYSE
– Great expansion of securities traded
– Rapid increase in sales activity
– Marked rise of volatility
• High variability around these trends in market ‘quality’
• Dramatic increase in average illiquidity
– Because the NYSE was expanding to serve a riskier population?
– Spurring economic growth?
– Planting the seeds of instability?
Liquidity measures
• Quoted spreads versus realized/effective spreads
– quoted spreads are observable
– realized/effective spreads are not observable – they are
due to price-improvement (i.e. negotiations)
Liquidity measures
• Basic effective spread measure (Roll, 1984)
• In efficient markets, variation in transactions prices results from
randomness of buy and sell orders plus positive transaction costs
•
Bid-ask bounce induces negative serial correlation in price changes
• With high transaction costs, deviation of transaction prices from
fundamentals not immediately arbitraged (even in efficient markets),
causing higher auto-covariance
rit =
siR = 2
pit − pit −1
pit −1
− cov(rit , rit −1 )
• GKN-measure (1991): corrects Roll measure in case of positively
correlated transaction returns
Liquidity measures
• In thin markets, price effects of individual trades are more
pronounced.
• With high transaction costs, deviation of transaction prices from
fundamentals not immediately arbitraged (even in efficient markets)
• Covariance of successive price changes provides information about
effective transaction costs
• Statistical properties of Roll measure:
• Downward biased (George, Kaul, Nimalendran, RFS, 1991)
• Slow convergence (Harris, JF, 1990)
Liquidity measures
• Amihud stock illiquidity measure
• average ratio of the daily absolute return to the dollar volume on that day
Aill t =
i
rti
∑ vol
t∈T
i
t
• T defines averaging period (monthly or annual).
Liquidity measures
• Bid-ask spread comprised of several components
(according to Huang, Stoll, RFS 1997)
– Inventory holding costs
– Adverse selection
– Order processing costs
• technological and legal/regulatory costs of processing a trade
• market power
• Infer order processing component from cross-section regression of
effective spread (GKN measure) on quoted spread
– Coefficient of quoted spread = order-processing component