Trade, Merchants and Lost Cities of the Bronze Age∗
Gojko Barjamovica , Thomas Chaneyb , A. Kerem Coşarc , and Ali Hortacsud
a
Harvard University, NELC
Toulouse School of Economics, and CEPR
c
Stockholm School of Economics, and CEPR
d
University of Chicago, Department of Economics, and NBER
b
January 24, 2016
(Preliminary and Incomplete, Comments Welcome)
Abstract
We analyze a dataset of business documents inscribed in the cuneiform script and
preserved on clay tablets informing us about the long-distance caravan trading system
that connected northern Iraq, northern Syria and central Turkey in 19th century BCE.
The standard gravity specification fits this ancient trade data remarkably well. Using
data on modern-day trade of Turkish cities with Iraq and Syria, we find persistent
patterns across four millennia. An inverse-gravity exercise suggests a methodology for
estimating the location of lost cities. The size distribution of merchant coalitions is
approximately Pareto.
1
Introduction
This paper analyzes a vast collection of commercial records from the earliest documented
and market-oriented long-distance trade in world history: the old Assyrian trade network
connecting northern Iraq, northern Syria and central Turkey in 19th century BCE. The
clay tablets on which the merchants inscribed their shipment consignments, expenses, and
contracts—excavated, translated and published by archeologists and historians for more than
a century—paint a rich picture of an exchange economy.
∗
This research is supported by the University of Chicago Neubauer Collegium for Culture and Society.
Correspondence: [email protected], [email protected], [email protected],
[email protected].
Using a digitized sub-sample, we investigate spatial patterns of trade and the structure of
merchants’ networks. The results reveal remarkable similarities to the patterns documented
from modern datasets. A gravity equation on a measure of trade relationships between
ancient cities yields a distance elasticity of trade close to negative unity. Second, the size
distribution of merchant coalitions (precursor of firms) is approximately Pareto.
While some ancient cities have been discovered and excavated, some have been lost to
humanity. Analyzing the texts for descriptions of trade routes connecting the cities and
the landscapes surrounding them, historians have developed qualitative conjectures about
potential locations of several lost cities. We demonstrate an alternative, quantitative methodology based on maximizing the fit of the gravity equation. In the first stage, we estimate
the distance elasticity of trade between a sub-sample of known cities. This yields the abovementioned estimate that is close to negative unity. In the second stage, we maximize the fit
of the gravity equation, parameterized with this coefficient, to the entire sample of known
and unknown cities. This yields predictions for the coordinates of lost cities. In this paper,
we apply the methodology to Purushaddum, a seemingly prominent yet lost Anatolian city,
and compare the qualitative and quantitative conjectures for its location.
2
Historical Background and Data
Our data comes from a collection of around 23,500 texts excavated at the archaelogical site
of Kültepe, ancient Kaneš, located in Turkey’s central Anatolian province of Kayseri. These
texts were inscribed on clay tablets in the Akkadian language in cuneiform script by ancient
Assyrian merchants, their families and business partners. The texts date to a period roughly
between 1930 and 1718 BCE, with around 90% of the sample belonging to just one generation
of traders, c. 1895 - 1865 BCE.
Originating from the city of Assur in modern-day northern Iraq, a few hundred Assyrian
merchants settled in Kaneš on a permanent or temporary basis, and maintained smaller
2
expatriate trading settlements in some 30 urban centres on the central Turkish plateau. In
total, an estimated 1000 Assyrian agents were settled in Anatolia at any given point during
the period in question. Kaneš was the regional hub of the overland commodity trade involving
the import of luxury fabrics and tin from Assur to copper rich Anatolia in exchange for silver
and gold. Assyrian merchants were also involved in supplying tin to other Anatolian city
states from Kaneš. Map 1 shows major cities and the trade routes of the period.
Most of the texts under consideration are commercial: business letters, shipment documents, accounting records and contracts. Fittingly, the tablets they were inscribed on
were found in merchants’ houses and archives. In a typical shipment document or expense
account, a merchant would inform its partner about the cargo and related expenses:
In accordance with your message about the 300 kg of copper, we hired some
Kanesites here and they will bring it to you in a wagon...Pay in all 21 shekels
of silver to the Kanesite transporters. 3 bags of copper are under your
seal...Here, Puzur-Assur spent 5 minas of copper for their food. We paid 5 2/3
minas of copper for the wagon.
Kt 92/k 313 (lines 4-8,14-22)
Occasional business letters facilitated information exchange about market and transport
conditions:
Since there is a transporter and the roads are dangerous, I have not led the
shipment to Hutka. When the road is free and the first caravan arrived safely
here, I will send Hutka with silver.
POAT 28 (lines 3-7)
While the actual cuneiform tablets are scattered all around the world in collections and
museums, around 10,000 texts have been transliterated into Latin alphabet, published in
various volumes and recently digitized by historians. In this draft, we use quantitative
information about cities and merchants mentioned in a sub-sample of 5,464 digitized texts
3
available to us. The texts also contain information about prices, financial contracts and
resolution of legal disputes, which we hope to analyze in future work.
The version of the data used in this draft (tabulated by Barjamovic 2011) contains 51
unique cities mentioned in texts, with 34 having some information on their locations. Only
a small number of these 34 cities, however, have been precisely located. 14 cities have either
been excavated or strong clues exist for associating them with a current place. Another 12
cities have been lost, but the texts contain clues about their locations. From analyses of
such textual evidence and the geography of the landscape, historians developed competing
theories about potential sites. Finally, the locations of 8 cities are mostly speculative. In our
empirical investigation, we label these three types of cities having location certainty from 1
to 3, ranked in terms of increasing certainty.
To construct a measure of bilateral commercial interactions between cities, we use dyadic
counts of joint attestations to city names in tablets. For instance, the number of texts in
which any two cities i and j are mentioned in a direct relationship, such as “I am traveling
from i to j,” is Xijdir . Some texts mention city pairs without a direct relationship such as “I
arrived from i yesterday and the caravan will leave for j tomorrow.” While these cases suggest
some interaction between cities, they are plausibly less informative about trade relationships.
These counts of co-occurrences are denoted Xijindir . In our empirical analysis, we estimate a
gravity equation on both measures using bilateral distances, and propose a methodology to
estimate uncertain city locations.
To finance the fixed costs of trading and overcome transaction costs, old Assyrian merchants organized their trade as partnerships and coalitions within a tightly-knit ethnic network.1 A sub-sample of texts contain references to the merchants involved. In the absence
of surnames, it is challenging to disambiguate the names, i.e., making sure that Mr. PuzurAssur mentioned in different tablets is the same person. Historians meticulously analyze
1
These relationships resemble the case of 11th century Maghribi traders documented by Greif (1989,
1993). It is plausible that organizing trade in this fashion helped to alleviate agency costs through similar
mechanisms of repeated interaction and informal enforcement.
4
the texts using the information about the time periods, patronymic lineages (Puzur-Assur,
son of...) and other familial ties to identify the individuals.2 Once disambiguated, counts of
meaningful linkages between merchants (owning a joint caravan or letters addressed to each
other) give us information about the teams or coalitions. While not necessarily indicating
joint-stock companies, these coalitions resemble firms. In our empirical analysis, we analyze
the size distribution of merchant coalitions.
3
Empirical Analysis
3.1
Gravity Equation
We start by estimating a gravity equation on our data:
ln(Xij ) = zi + zj + δ ln(distij ) + ij ,
(1)
where distij is the geodesic distance between cities i, j and Xij is the count of direct or indirect
co-occurrences, depending on the specification. We only use cities with location certainty 1
or 2, which reduces the sample to 26 cities. We first estimate (1) with OLS, using positive
observations. Out of 325 possible dyadic combinations, 62 and 71 pairs have Xijdir > 0 and
Xijindir > 0, respectively. 54 pairs have both measures positive. Given this prevalence of
zeros, we also estimate a PPML specification by replacing the dependent variable with Xij
(Silva and Tenreyro, 2006).
Table 1 presents the results. Validating the qualitative nature of the data, the distance
elasticity in OLS estimation is significant for direct relationships only (columns 1 and 2).
PPML regression yields a significant coefficient for both measures (columns 3 and 4). To our
knowledge, this is the oldest evidence for the gravity except the work of Bossuyt et al. (2001)
who also estimate a gravity-like specification using citations from Babylonian tablets dating
2
We are thankful to Adam Anderson for sharing with us the current version of the dataset from his thesis
work at Harvard University.
5
back to the 21 century BCE. Our estimates are remarkably close to distances elasticities
estimated from modern-day trade data. This confirms the puzzling persistence of the distance
effect documented by Disdier and Head (2008) across four millennia.
3.2
Persistence of Economic Significance
To inquire whether the economic significance as captured by the fixed effect estimates zi show
persistence over time, we first map the ancient cities to modern day provinces in Turkey.
Figure 2 plots the fixed effects against the (log) population of these provinces, which shows
a positive correlation. We then use the fixed effects estimated from the 2003-2012 trade of
these provinces with Iraq and Syria (Cosar and Demir, 2016). Figure 3 plots the ancient and
modern fixed effects. The correlation is 0.36, which is preliminary evidence for persistent
economic significance of locations over four millennia as it pertains to comparable trade
relationships.3
3.3
Inverse Gravity
Treating bilateral distances as an independent variable, the gravity equation helps to estimate
the elasticity of trade to distance. We now take the reverse approach: using the distance
elasticity estimated from trade between cities with high location certainty, we estimate the
geographic coordinates of a major city, Purushaddum, which has location certainty equal
to 2 in our sample. In our gravity estimation above, we used the location suggested by
Barjamovic (2011), who locates it around 300 km to the west of Forlanini (2008) in central
Anatolia (figure 4).
In the gravity setting, each city is identified by its size and coordinates (lati , longi ).
Repeating the estimation presented in column 1 of table 1 by excluding Purushaddum, we
obtain a statistically significant distance elasticity of -0.748 and city fixed effects zi . The
3
We also note that the city of Kaneš is 20 km away from the province of Kayseri, an important regional
commercial hub in Turkey. Without a doubt, its fertile land, central location and being the end of passages
coming from the Middle East imply strong location advantages.
6
fixed effect zp and coordinates (latp , longp ) of Purushaddum are unknown. We do, however,
observe its trade Xip > 0 with 11 of these 25 cities. If these relationships obey gravity with
some error ip , one can write 11 equations
ln(Xip ) = zi + zp − 0.748 ln(distip ) + ip ,
(2)
where the distance between i and Purushaddum is given by the Haversine formula:
s
distip = 2r · arcsin
sin2
lati − latp
2
+ cos(lati ) cos(latp ) sin2
longi − longp
2
!
.
r is the radius of the Earth at the poles (6356 km) or at the equator (6378 km).
We estimate Purushaddum’s coordinates by maximizing the fit of the gravity equation
to its trade relations. In particular, we solve
ˆ p , long
ˆ ) =
(ẑp , lat
p
argmin
11
X
2ip ,
(zp ,latp ,longp ) i=1
subject to equation (2) and the constraint that Purushaddum should be located to the west
of Kanes.4 The gravity-implied location plotted in figure 4 is closer to the area proposed by
Barjamovic (2011) based on qualitative textual evidence.
3.4
Size Distributions of Merchants, Teams and Coalitions
We now turn to a preliminary analysis of the distributional properties of the sizes of merchants, of teams and of coalitions.
Figure 5 plots the counter-cumulative distribution of merchant sizes in a log-log scale.
The distribution of merchant sizes appears to be approximately log-normal, although we
have not applied an explicit statistical test of that hypothesis for the moment.
4
The methodology resembles the trilateration of locations by global positioning systems. An early
application is by Tobler and Wineburg (1971) who imposed a quadratic distance elasticity.
7
Figure 6 plots the counter-cumulative distribution of team sizes in a log-log scale. The
distribution of team sizes appears to converge to a Pareto distribution (a straight line in
a log-log scale) for teams above size 6. We formally estimate the shape parameter of this
Pareto distribution below.
Figure 7 plots the counter-cumulative distribution of coalition sizes in a log-log-scale.
Again, the distribution appears to converge to a Pareto distribution for coalitions above size
6.
We now turn to a formal statistical test whether those size distributions can be approximated by a Pareto distribution, defined as,
Pr [Sizei ≥ S] ∝ S −β
where β is the shape parameter of the Pareto distribution. Formally, we estimate by OLS
the following rank-size relation, using the procedure in Gabaix and Ibragimov (2011),
1
= α − β ln (Sizei ) + i .
ln Ranki −
2
(3)
The results are presented in table 2. When data on all sizes are used, surprisingly, the
distribution of merchant, team and coalition sizes seems to be close to Zipf’s law. This result
however is misleading, as the above figure show. When only data on the upper tail of those
distributions are used, the distribution of team and coalition sizes are precisely approximated
by a Pareto distribution with shape parameter -3. This tail index is stable to truncating
the distribution anywhere above size 6. For the distribution of merchant sizes, the tail index
is not stable, and keeps increasing (in absolute terms) as the distribution is truncated at a
higher point.
The robust finding of a Pareto distribution with tail index -3 for team and coalition sizes
suggests that a strong empirical regularity is present.
8
References
Barjamovic, G. (2011). A Historical Geography of Anatolia in the Old Assyrian Colony Period.
Museum Tusculanum Press.
Bossuyt, A., L. Broze, and V. Ginsburgh (2001). On invisible trade relations between mesopotamian
cities during the third millennium bc. The Professional Geographer 53 (3), 374–383.
Cosar, A. K. and B. Demir (2016). Domestic road infrastructure and international trade: Evidence
from turkey. Journal of Development Economics 118, 232 – 244.
Disdier, A.-C. and K. Head (2008). The puzzling persistence of the distance effect on bilateral
trade. The Review of Economics and statistics 90 (1), 37–48.
Forlanini, M. (2008). The central provinces of hatti. an updating. In K. Strobel (Ed.), New
Perspectives on the Historical Geography and Topography of Anatolia in the II and I Millennium
BC, Number 1, pp. 145–188. LoGisma Editore.
Gabaix, X. and R. Ibragimov (2011). Rank- 1/2: a simple way to improve the ols estimation of
tail exponents. Journal of Business & Economic Statistics 29 (1), 24–39.
Greif, A. (1989). Reputation and coalitions in medieval trade: evidence on the maghribi traders.
The journal of economic history 49 (04), 857–882.
Greif, A. (1993). Contract enforceability and economic institutions in early trade: The maghribi
traders’ coalition. The American economic review , 525–548.
Silva, J. S. and S. Tenreyro (2006). The log of gravity. The Review of Economics and statistics 88 (4),
641–658.
Tobler, W. and S. Wineburg (1971). A cappadocian speculation. Nature 231, 39–41.
9
Appendix: Figures and Tables
Figure 1: Cities, Routes and Trade Patterns
10
Figure 2: Ancient Scale of Trade and Modern Population across Cities
Figure 3: Ancient and Modern Scales of Trade across Cities
11
Figure 4: Location of Purushaddum
Kanes
P_Barjamovic
P_Forlanini
P_Gravity
Figure 5: The Distribution of Merchant Sizes
Notes: All 2,748 merchants who have at least one trading partner. A merchant’s “size” is defined as the
number of unique trading partners of that merchant. The graph plots in a log-log scale the counter-cumulative
distribution of merchant sizes. Data sources are Old Assyrian period tablets from various archives.
12
Figure 6: The Distribution of Team Sizes
Notes: All 5,778 teams of merchants. A team is defined as a group of merchants mentioned in the same
text and who are in an explicit trading relationship. Team size is defined as the number of merchants in the
team. The graph plots in a log-log scale the counter-cumulative distribution of team sizes. Data sources are
Old Assyrian period tablets from various archives.
Figure 7: The Distribution of Coalition Sizes
Notes: All 4,061 coalitions of merchants. A coalition is defined as the union of teams, where two teams belong
to the same coalition if and only if at least one unique merchant belongs to both. Coalition size is defined as
the number of unique merchants in the coalition. The graph plots in a log-log scale the counter-cumulative
distribution of coalition sizes. Data sources are Old Assyrian period tablets from various archives.
13
Table 1: Gravity Estimates
ln(dist)
Method
Relationship
N. Obs.
R2
(1)
ln(X dir )
-0.775∗∗∗
(0.170)
OLS
Direct
62
0.704
(2)
ln(X indir )
-0.180
(0.119)
OLS
Indirect
71
0.732
(3)
X dir
-1.303∗∗∗
(0.173)
PPML
Direct
195
0.746
(4)
X indir
-0.569∗∗∗
(0.127)
PPML
Indirect
191
0.847
Notes: Standard errors in parentheses. ∗∗∗ p < 0.01. Relationship denotes whether the dependent variable X count meaningful commercial relationships betweens cities (direct), or any co-occurrence of their
names in the same document (indirect).
Table 2: Testing Rank-Size Relationship
β
N. Obs.
R2
All
Merchant
Team
∗∗∗
1.20
1.12∗∗∗
(0.0038) (0.0079)
20,748
5,778
0.830
0.776
Coalition
1.24∗∗∗
(0.012)
4,061
0.722
Size 6 and above
Merchant
Team
Coalition
∗∗∗
∗∗∗
1.98
3.11
3.19∗∗∗
(0.0032) (0.0054) (0.0043)
6,810
916
857
0.982
0.997
0.998
Notes: Standard errors in parentheses. ∗∗∗ p < 0.01. This table estimates equation 3 using
data on merchant, team and coalition sizes.
14