(PDF) Estimating the Effects of the Container Revolution on World Trade

1

Estimating the effects of the container revolution on world trade1

Daniel M. Bernhofen

American University, CESifo and GEP

Zouheir El-Sahli

Lund University

Richard Kneller

University of Nottingham, CESifo and GEP

August 26, 2015

Abstract

Many historical accounts have asserted that containerization triggered complementary technological

and organizational changes that revolutionized global freight transport. We are the first to suggest an

identification strategy for estimating the effects of the container revolution on world trade. Our

empirical strategy exploits time and cross-sectional variation in countries’ first adoption of container

facilities and combines it with product-level variation in containerizability and container usage.

Applying our container variables on a large panel of product level trade flows for the period 1962-

1990, our estimates suggest economically large concurrent and cumulative effects of containerization

and lend support for the view of containerization being a driver of 20th century economic

globalization.

JEL classification: F13

Keywords: containerization, 20th century global transportation infrastructure, growth of world trade.

1Address for Correspondence: Daniel M. Bernhofen, School of International Service, American University, 4400

Massachusetts Ave NW, Washington DC 20016-8071, Phone: 202-885-6721, Fax: 202-885-2494, email:

dbernhof@american.edu.

2

1. Introduction

One of the most striking developments in the global economy since World War II has been

the tremendous growth in international trade. As shown in Figure 1, the increase in world trade

accelerated dramatically during the early 1970s, with world trade growing in real terms from 0.45

trillion dollars in the early 1960s to 3.4 trillion dollars in 1990, by about a factor of 7. A central

question is what accounts for this dramatic growth in world trade. Two broad explanations have

been identified: (i) trade policy liberalization and (ii) technology-led declines in transportation costs.2

Figure 1: The growth of world trade (deflated): 1948-1990

A vast literature on transportation economics has argued that containerization was the major

change in 20th century transportation technology responsible for the acceleration of the globalization

of the world economy since the 1960s.3 Figure 1 reveals that the dramatic increase in the growth in

world trade appear to have coincided with the genesis of the global container era which can be dated

2 See Krugman (1995) for a prominent discussion on the growth of world trade.

3See Levinsohn (2006) and Donovan and Booney (2006) for good overviews of containerization and references to case

studies on the effects of containerization from a business history perspective.

3

to 1966. However, a quantitative assessment on the effect of containerization on international trade

appears to be missing. In fact, in an influential and well-researched book on the history of the

container revolution, Mark Levinson (2006, p.8) asserts that "how much the container matters to the

world economy is impossible to quantify". Our paper challenges this claim and suggests an

empirical identification strategy that allows us to estimate the effect of containerization on

international trade.

Containerization was invented and first commercially implemented in the US in the mid

1950s. After ten years of US innovation in port and container ship technologies, followed by the

international standardization in 1965, the adoption of containerization in international trade started in

1966. Numerous case studies have documented that containerization has not only affected the

operation and relocation of ports but the entire transportation industry.4 Specifically, the introduction

of containerization has gone hand in hand with the creation of the modern intermodal transport

system, facilitating dramatic increases in shipping capacities and reductions in delivery times through

intermodal cargo movements between ships, trains and trucks.

Based on information scattered in transportation industry journals, we are able to identify the

year in which a country entered the container age by first processing cargo via port and railway

container facilities. Since the intermodal feature of the container technology required a

transformation of the entire journey from the factory gate to the customer, we capture

containerization as a country-pair specific qualitative technology variable that switches from 0 to 1

when both countries entered the container age at time t. Time and cross-sectional variation of this

technology variable permits us to apply it to a large panel of bilateral trade flows for 157 countries

during the time period of 1962-1990. Our time horizon includes 4 years of pre-container shipping in

international trade, the period of global container adoption 1966-1983 and 7 years where no new

country in our sample started to adopt containerization. Since our time horizon precedes the period

of dramatic reductions in the costs of air transport, our study excludes the other major 20th century

change in the global transportation sector. Because our data provides information on both port and

railway containerization, our analysis captures the main modes of international transport during this

period.5

Our empirical strategy exploits cross-sectional and time variation in the economy-wide

adoption of container facilities and product level variation in containerizability in a difference-in-

difference framework. The inclusion of country-and-time effects allows us to capture multi-lateral

4 See McKinsey, (1967, 1972) and the various issues in Containerization International (1970-1992).

5 Since the adjustment of container transportation via truck followed the adoption of port and railway container facilities

we capture the main modes of cargo transport during this period.

4

resistance identified by the structural gravity literature and other time-varying factors that might be

correlated with countries’ decisions to invest in container ports6. Difficult to measure geographic

factors, like government desires to act as container port hubs, are captured by country-pair specific

fixed effects. The panel nature of our data set permits us not only to estimate the cumulate average

treatment effects (ATE) of containerization but also allows us to evaluate the size of the estimates in

comparison to the time-varying trade policy liberalization variables that have been used in the

literature.

Although the introduction of containerization resulted in economy-wide changes of

transportation infrastructure, the impacts are expected to be uneven among traded product groups.

For this reason and because not all products are containerizable, we examine variations in bilateral

trade flows at a disaggregated level. This allows us to exploit product level variations in

containerizability and container usage amongst adopters of the container. We exploit a 1968 study by

the German Engineers Society which classifies 4-digit product groups as to whether they were

suitable for container shipments as of 1968 and use also more recent US Census product level data

on container usage.

Restricting our sample to North-North trade, which are mainly the early adopters, our

benchmark specification which uses differences in the timing of adoption between countries suggests

that the cumulative average treatment effect (ATE) of containerization was about 1,240% after 15-

years. For all countries we find an effect that is smaller but still of economic importance at 900%.

Although the contemporaneous effect of containerization is quite similar to the North-North analysis,

the dynamic effects of containerization are much weaker for trade flows that involve developing

economies. This might be explained by the reduced penetration of the technology to include other

parts of the transport system in developing countries.

Interpreting the results from the above regressions relies on trade between non-containerized

pairs of countries providing a valid counterfactual. When testing this using information on trade

flows in the pre-containerization period, we find that this is satisfied for North-North trade in the

benchmark specification, but not consistently so across all of the robustness tests that we provide.

For the sample containing all countries (early and later adopters), we consistently find pre-container

differences in the growth rate of bilateral trade. Therefore, even though the literature suggests that

the returns to containerization were much more uncertain for the early adopters (North-North trade),

an identification strategy based solely on differences in the timing of adoption of the container does

6 See Feenstra (2004, p.161-163) for a good discussion on the use of country fixed effects to deal with multilateral

resistance.

5

not appear capable of providing convincing evidence of causal effects. The additional trade can be

attributed to the container revolution only with caution.

Those estimates may also be subject to endogeneity concerns; countries did not choose to

adopt this new transportation technology randomly and our estimates may be upward biased. To

address this we present a second set of estimates that exploits the containerizability of some

products. From this we continue to provide evidence that containerization did affect bilateral trade

flows, albeit more modestly compared to estimates based on differences in the timing of adoption.

Comparing the change in bilateral trade of containerizable products versus non-containerizable

products amongst pairs of countries that had adopted the container, we find significant differences in

the growth of trade in the order of 17.4% for North-North trade and 14.1% for World trade, but

where the latter occur 10-15 years after containerization. Comparing these to the estimates derived

purely from differences in the timing of adoption indicates that container was more important as an

explanation of differences in the growth of trade between countries across time, rather than across

products. Pre-containerization differences in trade flows are however less apparent when using

differences in trade flows in the product dimension, increasing the likelihood that these estimates are

causal.

Our paper contributes to the broader literature that aims to quantify the effects of changes in

transportation technology on economic activity. Starting with Fogel’s (1964) pioneering study on

the effects of US railroads on economic growth, a number of studies have investigated the effects of

railroad construction on economic performance and market integration. Based on detailed archival

data from colonial India, Donaldson (2012) provides a comprehensive general equilibrium analysis

of the impacts resulting from the expansion of India’s railroad network during 1853-1930.7 While the

introduction of rail and steamships were the main changes in transportation technology that

underpinned the first wave of globalization (1840s-1914), students of transportation technology and

prominent commentators link the post World War II growth of world trade to containerization. For

example, Paul Krugman writes (2009, p. 7):

“The ability to ship things long distances fairly cheaply has been there since the steamship and

the railroad. What was the big bottleneck was getting things on and off the ships. A large part of

the costs of international trade was taking the cargo off the ship, sorting it out, and dealing with

the pilferage that always took place along the way. So, the first big thing that changed was the

introduction of the container. When we think about technology that changed the world, we think

7 Donaldson (2012) tests several hypotheses of the effects of railroads that he derives from a multi-region, multi-

commodity Ricardian trade model. Hurd (1975) follows Fogel (1964) in applying a social savings methodology to

estimate the impacts of Indian railroad construction.

6

about glamorous things like the internet. But if you try to figure out what happened to world

trade, there is a really strong case to be made that it was the container, which could be hauled

off a ship and put onto a truck or a train and moved on. It used to be the case that ports were

places with thousands and thousands of longshoremen milling around loading and unloading

ships. Now longshoremen are like something out of those science fiction movies in which people

have disappeared and been replaced by machines”.

The current state of the empirical trade literature supports the view that the decline in

transportation costs did not play a significant role in the growth of world trade. In an influential

paper studying the growth of world trade, Baier and Bergstrand (2001) have found that the reduction

in tariffs is more than three times as important as the decline in transportation costs in explaining the

growth of OECD trade between 1958-60 and 1986-88.8 In his survey of how changes in

transportation costs have affected international trade in the post world War II period, Hummels

(2007) has detected an actual increase in ocean shipping rates during 1974-84, a period after the

adoption of containerization in the US. Using commodity data on US trade flows, Hummels finds

that freight cost reductions from increasing an exporter’s share of containerized trade have been

eroded by the increase in fuel costs resulting from the 1970s hike in oil prices.9

Our identification strategy is rooted in our reading of the historical literature which suggests

that containerization resulted in far reaching complementary technological and organizational

changes in port and railway services that affected economies’ entire transportation sectors.10 In fact,

our findings confirm Hummels’ (2007, p. 144) intuition that “the real gains from containerization

might come from quality changes in transportation services…To the extent that these quality

improvements do not show up in measured price indices, the indices understate the value of the

technological change”. Our findings are also compatible with Yi (2003) who has stressed the role of

vertical specialization and disintegration of production as a major factor in explaining the growth of

world trade.11 Experts in transportation economics have emphasized repeatedly that the global

8 Because of data limitations, Baier and Bergstrand (2001, Table 1, p.14) use only a multi-lateral rather than a bilateral

index of changes in transportation costs. Their index suggests that Austria’s transportation costs versus the rest of the

world have actually increased between 1958 and 1986, which does not appear plausible. Although land-locked, Austria

has been an early entrant in the container age through their construction of container railway terminals in 1968 which

connected it to the main container ports in Europe.

9 Another study that investigates the effects of containerization on US imports in the post adoption period is Blonigen and

Wilson (2008). Building on Clark, Dollar and Micco (2004), they estimate the effects of port efficiency measures on

bilateral trade flows and find that increasing the share of trade that is containerized by 1 percent lowers shipping costs by

only 0.05 percent.

10 A study by Eyre (1964) claims that during the pre-container age ocean shipping costs accounted only for 25% of the

total door-to-door costs of shipping a truckload of medicine from Chicago to Nancy in France.

11 Baier and Bergstrand (2001) are quite honest in pointing out that their final goods framework excludes this potential

source of the growth of world trade.

7

diffusion of intermodal transport was a prerequisite for the disintegration of production and the

establishment of global supply chains (Notteboom and Rodrigue, 2008).

The next section of the paper provides a historical discussion on the origins and effects of

containerization. Our historical narrative fulfills two purposes. First, by describing the different

channels through which container adoption reduced trade costs we point to the mechanisms that

appear to be responsible for our estimated effects. Second, our historical evidence on the speed of

diffusion of container technology within the transportation structure of two selected economies

provides the rationale for our identification strategy of capturing containerization. Section three

introduces our empirical specifications and discusses our empirical findings. Section four concludes.

2. The container revolution and intermodal transport

“Born of the need to reduce labor, time and handling, containerization links the manufacturer or

producer with the ultimate consumer or customer. By eliminating as many as 12 separate

handlings, containers minimize cargo loss or damage; speed delivery; reduce overall

expenditure”.

(Containerisation International, 1970, p. 19)

2.1 Historical background

Before the advent of containerization, the technology for unloading general cargo through the

process of break-bulk shipping had hardly changed since the Phoenicians traded along the coast of

the Mediterranean. The loading and unloading of individual items in barrels, sacks and wooden

crates from land transport to ship and back again on arrival was slow and labor-intensive.

Technological advances through the use of ropes for bundling timber and pallets for stacking and

transporting bags or sacks yielded some efficiency gains, but the handling of cargo was almost as

labor intensive after World War II as it was during the beginning of the Victorian age. From a

shipper's perspective, often two-thirds of a ship's productive time was spent in port causing port

congestion and low levels of ship utilization. Following the spread of the railways, it became

apparent already during the first era of globalization that the bottleneck in freight transport was at the

interface between the land and sea transport modes.

Before World War II, US, British and French railway companies experimented with methods

of sealing goods in different sizes and shapes of boxes before transporting them. However, the lack

of specialized capital equipment like specialized cranes for loading and unloading combined with

union resistance to changes in work practices at the docks delayed the development of container

shipping until the mid-1950s.

8

The genesis of the container revolution goes back to April 26, 1956 when the Ideal- X, made

its maiden voyage from Port Newark to Houston, Texas. The ideal X was a converted World War II

tanker that was redesigned with a reinforced deck to sustain the load of 58 containers. As so

common in the history of innovation, the breakthrough of containerized shipping came from

someone outside the industry, Malcom McLean, a trucking entrepreneur from North Carolina.

Concerned about increased US highway congestion in the 1950s when US coastwise shipping was

widely seen as an unprofitable business, McLean's central idea was to integrate coastwise shipping

with his trucking business in an era where trucking and shipping were segmented industries. His

vision was the creation of an integrated transportation system that moved cargo door to door directly

from the producer to the customer. The immediate success of the first US container journey resulted

from the large cost savings from the mechanized loading and unloading of containerized cargos.

Shortly after the Ideal-X docked at the Port of Houston, McLean's enterprise, which later became

known as Sea-Land Service, was already taking orders to ship containerized cargo back to Newark.

The 1956 container operation by the Ideal X involved a ship and cranes that were designed

for other purposes. McLean's fundamental insight, which was years ahead of his time, was that the

success of the container did not rest simply on the idea of putting cargo into a metal box. Instead, it

required complementary changes in cranes, ships, ports, trucks, trains and storage facilities. Three

years following Ideal X's maiden voyage, container shipping saw additional savings through the

building of purpose-built container cranes followed by the building of large purpose-built

containerships. On January 9, 1959 the world's first purpose-built container crane started to operate

and was capable of loading one 40,000-pound box every three minutes. The productivity gains from

using this container crane were staggering, as it could handle 400 tons per hours, more than 40 times

the average productivity of a longshore gang.12 Investment in larger shipping capacity became now

profitable since containerization dramatically reduced a ship's average time in ports.

Given the large investment costs, industry experts revealed a considerable amount of

uncertainty and skepticism regarding the success of the container technology at the time. Many

transportation analysts judged container shipping as a niche technology and did not anticipate the

dramatic transformations that this technology was about to bring to the entire domestic and

international transportation sector. In the first decade following the Ideal- X’s maiden voyage,

innovation and investment in container technology remained an American affair. But, as Levinson

(2006, p. 201) points out, "ports, railroads, governments, and trade unions around the world spent

those years studying the ways that containerization had shaken freight transportation in the United

12 See Levinson (2006, p. 65).

9

States". The early initiatives came from US shipping lines and by the early 1960s, containerization

was firmly established on routes between the US mainland and Puerto Rico, Hawaii and Alaska. Ten

years of US advancement in container technology set the foundation for containerization to go global

in 1966.13 In that year, the first container services were established in the transatlantic trade between

the US and European ports in the UK, Netherlands and West Germany.

2.2 Economic effects of containerization

From a transportation technology perspective, containerization resulted in the introduction of

intermodal freight transport, since the shipment of a container can use multiple modes of

transportation -ship, rail or truck- without any handling of the freight when changing modes. By

eliminating sometimes as many as a dozen separate handlings of the cargo, the container resulted in

linking the producer directly to the customer. Since containerization resulted in a reduction of the

total resource costs of shipping a good from the (inland) manufacturer to the (inland) customer, its

impact is not adequately captured by looking at changes in port-to-port freight costs.14

Containerization started as a private endeavor by the shipping lines. In the early stages,

shipping lines had to bear most of the costs since many ports such as New York and London were

reluctant to spend significant funds on ‘a new technology’ with uncertain returns at the time. Many

shipping lines had to operate from small and formerly unknown ports and install their own cranes.

The process was extremely expensive. After the container proved to be successful, ports warmed up

to containerization and a race started among ports to attract the most shipping lines by building new

terminals and providing the infrastructure to handle containers. Containerization required major

technological changes in port facilities, which often led to the creation of new container ports. In the

United States, the new container ports in Newark and Oakland took business from traditional ports

like New York and San Francisco. In the UK, the ports of London and Liverpool, which handled

most of the British trade for centuries, lost their dominant position to the emerging container ports of

Tilbury and Felixstowe.

In many countries, port authorities fall under the administration of the government. Because

of the high costs, careful planning and analysis had to be undertaken by governments to study the

13 Australia was the first country to follow the US and adopted container technology in 1964, but not in international

trade.

14 Reliable data on comparable changes in door-to-door and ocean freight rates before and after containerization are not

available. However, Eyre (1964) uses data from the American Association of Port Authorities to illustrate the

composition of estimated door-to-door costs of shipping one truckload of Medicine from Chicago to Nancy (France) in

the pre-container age. Astonishingly, ocean shipping amounted only 24.4% of total costs, whereas total port costs

constituted 48.7%, freight to the US port city 14.3%, European inland freight 8.6% and local freight in port vicinity 4%.

This supports the view that the bulk of costs savings from containerization stemmed from efficiency gains in the sea-land

interface.

10

feasibility of containerization. In the UK, the government commissioned McKinsey (1967) to

conduct a cost and benefit analysis before spending significant public funds on container port

facilities. Five years later, McKinsey (1972) provided a quantitative assessment of the effects of

containerization following the first five years after its adoption in the UK and Western Europe. Table

1 provides a summary of the sources and magnitude of resource savings from the adoption of

container technology between 1965 and 1970/71.

Table 1: Effects of containerization (UK/Europe)

Pre-container:

1965

Container:

1970/71

Productivity of dock labor

1.7 (tons per hour)

30 (tons per hour)

Average ship size

8.4 (average GRT*)

19.7 (average

GRT*)

Port concentration

(number of European loading ports,

southbound Australia)

11 ports

3 ports

Insurance costs

(Australia-Europe trade for imports)

£0.24 per ton

£0.04 per ton

Capital locked up as inventory in transit

(Route: Hamburg-Sydney)

£2 per ton

£1 per ton

Source: Authors' own compilation from various sources in McKinsey (1972).

*GRT is gross registered tonnage which is a ship's total internal volume expressed.

One of the major benefits of containerization was to remove the bottleneck in freight

transport in the crucial land-sea interface. The construction of purpose-designed container terminals

increased the productivity of dock labor from 1.7 to 30 tons per hour (Table 1). Improvement in the

efficiency and speed of cargo handling allowed shipping companies to take advantage of economies

of scale by more than doubling the average ship size. The resulting increase in port capacity provided

opportunities and pressures for the inland distribution of maritime containers. In the UK the

introduction of railway container terminals went in tandem with port containerization and by 1972

the Far East service alone already operated trains between an ocean terminal and six inland rail

terminals.

In the pre-container age, port managers handled and organized the trade of their own

industrial hinterlands. With the railways taking over the inland distribution, containerization

11

eliminated the notion of a port hinterland and containerized freight became concentrated in a few

major terminals. For example, whereas in 1965 ships in the (southbound) Australian trade called at

any of 11 loading ports in Europe, by 1972 the entire trade was shared among the three ports of

Hamburg, Rotterdam and Tilbury. Within a few years, a hub-and-spoke system already emerged.

A major benefit of sealing cargo at the location of production in a box to be opened at the

final destination is that it reduced the pilferage, damage and theft that were so common in the age of

break-bulk shipping. A common joke at the New York piers was that the dockers’ wages were

“twenty dollars a day and all the Scotch you could carry home”. The resulting reduction in insurance

costs from containerization was considerable. On the Australia-Europe trade, between 1965 and

1970/71 the insurance costs fell from an average of 24 pennies per ton to 4 pennies per ton (Table 1).

Intermodal transport also decreased the time in transit between cargo closing and availability.

Containerization cut the journey between Europe and Australia from 70 to 34 days. Given that the

average cargo at the time was worth about £60 per ton and assuming that the opportunity cost of

capital tied up in transit is about 15%, the 36-day improvement cut the capital cost of inventory by

about a half (Table 1).

The importance of labor in the operation of ports in the pre-containerization age resulted in

the emergence of strong labor unions, which resisted not only labor-saving organizational changes

but were also well-organized and effective in calling for strikes. The replacement of capital for labor,

which emerged through containerization, ended the frequent delays and uncertainties in shipping

caused by these strikes.15

2.3 Diffusion of container technology

The early use of containers was driven by private shipping companies who used container

sizes and loading devices that best fit their cargo and shipment routes. The first fully containerized

ships used 35-foot containers, which was the maximum allowable length for truck traffic on US

highways. However, since a fully loaded container of this size was too heavy for a crane to lift, other

companies used much smaller sizes which could be much easier stacked and moved with forklifts. A

major force for the international adoption and diffusion of container technology was the

standardization of container sizes. The standardization process was initiated in the US by the Federal

Maritime Board and involved stakeholders from the maritime sector, truck lines, railroads and trailer

manufacturers. In 1961, the Federal Maritime Board established the standard nominal dimension of

containers - 8 feet wide, 8 feet high and 10, 20, 30 and 40 feet long- and announced that only

15 We thank David Smith, Economics Editor of the Sunday Times, for pointing out the elimination of strikes as a channel

for cost reductions from containerization.

12

containerships designed for these sizes were able to receive construction subsidies from the US

government.

Following the setting of standards in the US, the International Standards Organization (ISO)

started to study containerization with the purpose of establishing worldwide guidelines as a

prerequisite for firms and governments investing in internationally compatible container technology.

Following a compromise between US and European interest groups, the ISO formally adopted the

10- ,20- , 30- and 40-foot containers plus a few smaller sizes favored by the Europeans as ISO

standards in 1964. Besides container size, strength requirements and lifting standards were other

major aspects the ISO was able to standardize in 1965. The standardization of container technology

was followed by the rise of container leasing companies who had now the incentives to expand their

fleets and allowing shippers the flexibility to lease containers and therefore significantly reducing the

fixed costs of using this technology. The ability for land and sea carriers to handle each other’s

containers in different locations set the foundation of global container adoption around the globe.

The adoption of container technology in international trade started in 1966 and the period of

1966-1983 has been labeled by geographers as the era of global diffusion of container technology

around the globe (Kuby and Reid, 1992).16 The introduction of container technology started with a

country’s investment in container port facilities but quickly progressed to engulf other parts of the

transportation network, like rail and road transport. Accompanying technological changes were

larger ships and trains and increased use of computers and telecommunications for managing and

tracking intermodal movement.

From our calculations of an underlying sample of 157 countries, 122 entered the container

age by first processing container cargo via port or railway facilities between 1966 and 1983 while 35

countries remained uncontainerized as of 1990. Appendix Table 3 lists all sample countries and

reveals considerable cross-sectional and time variation of countries’ adoption of container port

facilities during the sample period.

Historical accounts suggest an important role for the US military build-up in Vietnam to

stimulate the diffusion of container technology in East Asia. Initially, US military leaders were

skeptical about the adoption of container technology for the shipment of military cargo. However

Malcom McLean’s persistence and vision that his brainchild could alleviate the logistical challenges

associated with the rapid buildup of military forces in Vietnam after 1965 won the argument in favor

of the container. And by 1970 half of the military cargo was already containerized on critical

16 According to Kuby and Reid (1992, p.285) “…after 1982 the industry reached maturity, characterized by low margins

and greatly improved services….the containerization trend stabilized between 1985 and 1988, as near-saturation of the

new technology occurred.”

13

routes.17 A key feature of the military use of containerization during the Vietnam conflict was that

McLean’s ships were loaded with military freight on their west-bound journey to Vietnam while the

return east-bound journey carried initially empty containers. Levinson (2006) has argued that the

availability of this container capacity played a role in the diffusion of containerization.18

2.4 Capturing the degree of container utilization for two island economies

How quickly did containerization diffuse through an economy’s transportation sector

following its initial adoption of container port facilities? First, some tradable goods like assembled

automobiles, heavy machinery, construction equipment and some steel products can’t be put into

containers. Second, technological advancement in container technology has expanded the range of

containerizable products over time. For example, initially food products were not containerizable, but

through the development of refrigerated containers, food became containerizable in later years. A

measure of the degree of an economy’s container utilization in its international trade is the ratio of

the economy’s traded containerized cargo dived by its total trade of containerizable cargo. The

container utilization index for an economy i at time t, ContUtilit, is then given by:

ContUtilit = (containerized cargoit)/(containerizable cargoit) (1)

Fortunately, the denominator of (1) can be calculated based on a study by the Verband

deutscher Ingenieure (German Engineers Society) which classified 4-digit SITC industries according

to whether they were suitable for containers as of 1968.19 The calculation of (1) faces the challenge

that containers can cross borders through different modes of transport (sea, rail, truck and air) and

that there exist no data on traded container tonnage via rail, trucks or air. However, for island

economies like the UK and Japan, where (at the time) the majority of international trade went

through sea ports, it is possible to trace their growth of container utilization following their first

adoption of the technology.20 For these two countries, (1) can be calculated by combining data on

container tonnage going through seaports available from Containerisation International with tonnage

17 Levinson (2006), p. 183.

18 While working on this project, we became aware of the very interesting work by Gisela Rua (2014) who formally

investigates the determinants of the diffusion of containerization. She finds that the decision of later adopters is

influenced by their trade with the US. Her study is based on detailed data on port container usage, but does not focus on

containerizable versus non-containerizable products.

19 The engineering study classifies industries into Class A: suitable for containers, Class B: goods of limited suitability

for containers and Class C: goods not suitable for containers as of 1968. For the purpose of constructing our container

utilization index, we only include goods in Class A as containerizable. These are listed in Appendix Table 4.

20 Between 1965 and 1979, 99% of UK trade went through seaports.

14

of trade in containerizable industries provided by the OECD International Trade by Commodity

Statistics.

Figure 2 depicts the time change in the container utilization index (1) for the UK and Japan

during the period of 1965-1980.21 The UK adopted its first container facilities in 1966 and the

technology diffused quite rapidly. The utilization grows from 20% in 1967 to about 80% in 1973 and

remains then quite flat, which can be explained by the oil shock and the recession following the oil

crisis. The picture for Japan is quite similar. Japan adopted containerization in 1969 and the

utilization index grows from 20% in 1970 to about 50% in 1973 and then to over 80% during the late

1970s.22

Figure 2: Changes in container utilization in the UK and Japan

2.5 Containerization and land transport

A defining element of the adoption of container technology is the creation of intermodal

transport. Containerization was quickly picked up by railways in different countries. For example,

in response to port containerization, and in an effort to avoid being left out, the railways of Europe

21 Total trade is defined as (exports+imports)/2. Because of missing data in the years of container adoption, the graph

depicts linear segments between 1965 and 1967 for the UK and between 1968 and 1970 for Japan.

22 Rua’s (2014, p.36) documentation of a relatively lower container usage in her broad sample of countries in her Figure 1

can be reconciled by noting that we focus only on two ‘major adopters’ and that the denominator in our container

utilization index pertains to total trade in containerizable products rather than total trade.

15

came together in 1967 and formed Intercontainer, The International Association for Transcontainer

Traffic. This company was formed to handle containers on the Continent and compete with

traditional shipping lines.23 Railway containerization allowed landlocked countries like Austria and

Switzerland to ship their goods in containers to seaports in neighboring countries destined to

overseas destinations. In many cases, this was cheaper and less laborious than road transportation. In

a comprehensive cost study for the UK, McKinsey (1967) calculated that container transport was

cheaper by rail than truck for journeys above 100 miles. Containerisation International (1972)

estimated that the cost of moving 1 TEU (twenty foot equivalent unit container) between Paris and

Cologne in 1972 was about 75% of the equivalent road costs.24

Although the majority of non-land locked countries adopted railway container facility after

their introduction of container seaports, for some the ordering was reversed. For example, Norway

entered the container age via their railway network in 1969, five years before the adoption at their

seaports. This suggests that countries could enter the container age either through the introduction of

rail or sea ports container facilities.25

3. Empirical implementation

3.1 Quantifying containerization

Our objective is to estimate the effect of containerization on international trade. The key

question that arises is how to capture this technological change quantitatively. Since the adoption of

container technology triggered complementary technological and organizational changes that

affected an economy’s entire transportation system this suggests quantifying this technological

change at the economy level. If economy level data on the international shipments via rail or truck

were available, the container utilization index (1) would be a sensible measure of the technological

change. However, because of the absence of the appropriate data and the occurrence of

technological change regarding containerizability, we can’t go this path. Alternatively, the quick

rise in the container utilization for the UK and Japan justifies quantifying containerization by a

23 At the time, British Rail was already operating a cellular ship service between Harwich, Zeebrugge and Rotterdam and

a freightliner service between London and Paris. Initially 11 European countries formed Intercontainer and were later

joined by 8 more.

24 We are not aware of any studies that document the spread of road containerization equivalent to that for port or rail

containerization. The historical narrative suggests however that the developed countries that we focus on this was

concurrent to port and/or rail containerization. Outside of this the assumption is less certain. We found for example

photographic evidence that the Comoros Islands were using containers without either rail or port container depots (they

were instead dropped onto modified rowing boats).

25 Besides containerization, advances in information technologies and logistics, in particular ‘just in time’ manufacturing

played a second key role in facilitating the rise in trade and global supply chains. See Levinson (2006, chapter 14) for an

anecdotal discussion on the joint effects. Because the technological changes in logistics started to kick in only in the

1980s, it is beyond the scope of our paper.

16

qualitatively variable that switches from 0 to 1 when country i entered the container age at time t.

We define entrance into the container age as the first recorded use of either seaports or inland railway

ports. An advantage of this specification is that it captures the intermodal aspect of containerization

since it encompasses land-locked countries like Austria and Switzerland who entered the container

age via rail connection to the container seaports of Rotterdam and Hamburg. Based on the recorded

information provided in the published volumes of Containerisation International between 1970 and

1992, we construct a time-varying container adoption variable for country i, adoptcontit, defined as:

𝑎𝑑𝑜𝑝𝑡𝑐𝑜𝑛𝑡!"!=1!!𝑖𝑓!𝑐𝑜𝑢𝑛𝑡𝑟𝑦!𝑖!𝑢𝑠𝑒𝑠!𝑠𝑒𝑎 !𝑜𝑟!𝑟𝑎𝑖𝑙!𝑐𝑜𝑛𝑡𝑎𝑖𝑛𝑒𝑟!𝑝𝑜𝑟𝑡𝑠!𝑖𝑛!𝑦𝑒𝑎𝑟!𝑡!!!!!!!!!!!

!!!!!0!!!!!!!𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! (2)

Although a country’s adoption of container technology is expected to have some effect on its

overall trade, the nature of container technology suggests that containerization is more adequately

captured in a bilateral trading context since this allows us to specify the presence of container

technology of a country’s trading partner. This suggests capturing containerization by a time-

varying bilateral technology variable, defined as:

!!!!𝑐𝑜𝑛𝑡!"#!=1!!𝑖𝑓!𝑖!𝑎𝑛𝑑!𝑗!ℎ𝑎𝑣𝑒!𝑏𝑜𝑡ℎ!𝑐𝑜𝑛𝑡𝑎𝑖𝑛𝑒𝑟𝑖𝑧𝑒𝑑!𝑝𝑜𝑟𝑡𝑠!𝑜𝑟!𝑟𝑎𝑖𝑙𝑤𝑎𝑦𝑠!!𝑖𝑛!𝑦𝑒𝑎𝑟!!𝑡

!!0!!!!!!!!!!!!!!!!!!!!!!𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! (3)

From an econometric point of view, there are several advantages quantifying containerization

by a time-varying bilateral variable, most notably the ability to include various types of fixed effects.

Since we are interested in exploring causal effects of container technology on trade, one worries

about potential selection bias in the adoption of container technology. Armed with “ex post”

knowledge that containerization revolutionized global freight transport, one would be inclined to

infer that countries that initially traded a lot would be most likely to adopt container technology.

However, as we have pointed out in our historical narrative in section 2, from the relevant decision

point of view of the 1960s, containerization was by many viewed as a niche technology with little

perceived impact on the volume of international trade. For example, Levinson (2006, p. 276) writes:

“The huge increase in long-distance trade that came in the containers wake was foreseen by

no one. When he studied the role of freight in the New York region in the late 1950s, Harvard

economist Benjamin Chinitz predicted that containerization would favor metropolitan New York’s

industrial base by letting the region’s factories ship to the South more cheaply than plants in New

England or the Midwest. Apparel, the region’s biggest manufacturing sector, would not be affected

by changes in transport costs, because it was not ’transport-sensitive’. The possibility that falling

transport costs could decimate much of the U.S. manufacturing base by making it practical to ship

17

almost everything long distances simply did not occur to him. Chinitz was hardly alone in failing to

recognize the extent to which lower shipping costs would stimulate trade. Through the 1960s, study

after study projected the growth of containerization by assuming that existing export and import

trends would continue, with the cargo gradually shifted into containers”.

The second component of our identification strategy exploits the fact that not all products are

containerizable. Specifically, we take advantage of the 1968 study by the German Engineer’s society,

reported in Containerisation International Yearbook (1971), which classifies 4-digit product lines

whether they are suitable for container shipments as of 1968.26 Unfortunately, we are not aware of

any other study that provides a classification of product containerizability after 1968.27 However, the

Foreign Trade Statistics Division of the US Census Bureau collects port level data on the mode of

transportation of US imports and exports by products which goes back to 2003.28 Because the US

Census data provides information on whether imports entered the US via container vessels, it allows

us to provide adjustments on products that were categorized as containerizable or uncontainerizable

in 1968.

Based on these two data sources we construct three measures of containerizability. We use

the 1968 classification within our benchmark regressions, based on the arguments that this measure

is based on generic engineering criteria, while the US Census Bureau data capture revealed

containerizability; and that the date of this classification, 1968, is closer to the major adoption period,

which ran between 1966 and 1983. Results for the second and third measures can be found in the

robustness section and are constructed using the US Census Bureau data to either redefine the ‘non-

containerizable control group’ or both the treatment and the control group. The second measure uses

the US Census data to reduce the 1968 control group of ‘non-containerizables’ by excluding products

within this category with relatively high container usage in 2003.29 The third measure makes the

same adjustment to ‘non-containerizable’ products and in addition excludes from the category of

‘containerizable’ products those with low container usage in 2003.30 Containerizability of product

group k is captured by the categorical variable prodk, defined as:

𝑝𝑟𝑜𝑑!=1!𝑖𝑓!𝑘!𝑖𝑠!𝑐𝑙𝑎𝑠𝑠𝑖𝑓𝑖𝑒𝑑!𝑎𝑠!𝑐𝑜𝑛𝑡𝑎𝑖𝑛𝑒𝑟𝑖𝑧𝑎𝑏𝑙𝑒!𝑖𝑛!1968!𝑤𝑖𝑡ℎ𝑜𝑢𝑡/𝑤𝑖𝑡ℎ!2003!𝑎𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡

0!𝑖𝑓!𝑘!𝑖𝑠!𝑛𝑜𝑡!𝑐𝑜𝑛𝑡𝑎𝑖𝑛𝑒𝑟𝑖𝑧𝑎𝑏𝑙𝑒!𝑖𝑛!1968!𝑤𝑖𝑡ℎ𝑜𝑢𝑡/𝑤𝑖𝑡ℎ!2003!𝑎𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡 !!(4)

26 Recall that we used this classification in the construction of our utilization index (1) in Figure 2.

27 For example, technological progress expanded the scope of products that become containerizable, such as automobiles.

28 The data is accessible via USA trade online, categorized as Port Data (6-digit, HS level).

29 In the adjustment we ranked products according to the container usage ratio, (containerized trade in k)/(total trade in k).

We then dropped those products from the control group for which the container usage ratio was larger than in the bottom

quartile of the distribution. We chose 2003 because this is the earliest year for which this data is available.

30 We identify such products by calculating the share of US trade carried by container in 2003 and excluding from the

category of containerizable products those with a value in the bottom quartile of the distribution.

18

3.2 Research design and empirical specification

The time frame of our analysis is dictated by the availability of bilateral trade data at the

product level and the timeline of container adoption in international trade. Fortunately, the world

trade data set compiled by Feenstra et al. (2005) goes back to 1962 and covers bilateral trade flows

from 1962-2000 at the 4-digit product level.31 Since the adoption of containerization in international

trade started in 1966 and ended in 1983, we chose 1962-1990 as our sample period, which includes 4

years prior to the first adoption and 7 years past the last adoption year. We chose to exclude the

1990s because of both the redrawing of the political map after the end of the Cold War and the

reduction in the costs of air transport which started to kick in in the early 1990s. Although there is

limited data on changes in the mode of transport in international trade, a reading of the transportation

industry literature suggests that during our chosen sample period containerization was the main

technological change affecting the three major modes of transport (sea, rail and road) in international

trade.

Our outcome variable pertains to the bilateral trade flows between countries i and country j in

product group k at time t, 𝑥!"#$. Our empirical equation to be estimated is given by32:

∆ln 𝑥!"#$ =𝛽!+𝛽!∆𝑐𝑜𝑛𝑡!"# +𝛽!∆𝑐𝑜𝑛𝑡!"# ∙𝑝𝑟𝑜𝑑!+𝛽!∆𝑃𝑜𝑙𝚤𝑐𝑦!"# +!𝛽!𝐷!"#$ +!!𝑢!"#$. (5)

Our key explanatory variables are the time varying bilateral container variable contijt from (3) and the

interaction of this variable with the product containerizability variable prodk from (4). !𝑃𝑜𝑙𝚤𝑐𝑦!"#

denotes a vector of time-varying bilateral policy variables,!𝐷!"#$ includes a (large) vector of country-

time and product-time specific fixed effects and 𝑢!"#$ denotes the error term.

We opted for first differencing the data across our 5-year time periods such that our

dependent variable becomes Δxijkt=xijkt-xijk(t-1).33 Wooldridge (2002, chapter 10) suggests that first-

differencing a panel data set yields advantages if unobserved heterogeneity in trade flows is

correlated over time. So we regress Δxijkt on Δcontijkt and the other first differenced country-pair time-

variant policy variables like being in a regional trade agreement (rta) or a member in the GATT.

The time varying bilateral container variable Δcontijt captures the effect on the change in the

log of bilateral-product trade flows from differences in the timing of adoption of the container

31 The data set is constructed from the United Nations trade data and is available from NBER.

32 For efficiency of exposition, we omitted all lags of the explanatory variables.

33 We experimented with time periods of different lengths, but this has little effect on the main findings from the paper.

19

between bilateral pairs.34 The interaction of this with the containerizability variable allows for

differences in these effects for products that were suitable to be placed inside containers versus those

that were not. This difference compared to non-containerizable products might include reductions in

the time spent at port due to greater inter-modality of the logistics chain, reduced pilfering of goods,

reduced insurance costs etc. When using this measure of containerization the effects of

containerizability are measured relative to trade in non-containerizable products amongst country-

pairs that had adopted the container. In both cases, the anecdotal evidence suggests that

containerization led to the increased growth of trade and we therefore anticipate the coefficient on

these variables to be positive. The countrywide changes in transport infrastructure which occurred in

many countries when adopting the container would suggest there may be benefits for trade in all

types of products and therefore the estimate of β1 will be larger than β2.

Interpreting the coefficients on the various container measures as capturing causal effects

rests on the absence of omitted variables that are correlated with containerization and the error term

from the regression. Concern for the effects of such variables motivates the inclusion of various fixed

effects which we lay out below. It is possible that the effects of omitted variable bias remain

however; the presence of time-varying bilateral shocks may have led to increased bilateral trade

(faster growth), bringing forward in time the adoption of the container. We would over-estimate the

effects of the container on bilateral trade in such a case. The presence of such shocks represents the

main advantage to the inclusion of the containerizability interaction within equation (5). For the

containerizability interaction to be interpreted as capturing causal effects requires that these

unobservable shocks which determined the timing of adoption were not specific to containerizable

products. While it is not possible for us to rule out this possibility, it is much more challenging to

think what the omitted variables that affect trade in a sub-set of products that fitted inside containers

might be. An interpretation of causal effects may therefore be considered more convincing in such a

case.

An attractive feature of our panel specification is that it allows us to examine the dynamic

aspects of containerization over a time period 1962-1990 characterized, as argued above, by little

other transportation technological changes affecting international trade. Following the advice of the

panel literature (Wooldridge, 2002) we examine changes in trade flows at 5-year intervals. In our

context, the advantage of focusing on 5-year variations is that it mitigates the effect of differences in

the speed of adoption (recall Figure 2) as well as allowing time for the build-up of the intermodal

34 We also use this change in logs to calculate the growth in trade that occurred as a result of the container (equal to

x%=ecoeff-1) or the annual growth rate (equal to x%=e(coeff/t)-1).

20

transport system. We estimate equation (5) on 7 time periods: 1962, 1967, 1972, 1977, 1982, 1987

and 1990.

To our knowledge, (5) is the first specification in the literature that identifies a time-varying

bilateral technological change and aims to estimate its impact on international trade. An advantage of

this specification is that allows for a comparison to other time-varying bilateral policy changes for a

country pair i and j, like entrance into a regional trade agreement (rta) or both being a member in the

GATT, which have been treated extensively in the literature. Specifically, (5) allows for a

comparison between the estimate size of our container technology variable contijt and these trade

liberalization variables.35

An advantage of our panel specification is that it allows us to use fixed effects to avoid

omitted variable biases associated with multi-lateral resistance terms identified from the structural

approach to gravity.36 Specifically, the inclusion of time-varying importer (it), exporter (jt) and

product (kt) fixed effects in!𝐷!"#$ capture time-varying product and country-specific factors that are

either difficult to pin down or measure.37 In addition to multilateral resistance, the country-time

dummies will also capture other changes that effect the growth of trade flows of a country. These

include the infrastructure investments or other changes that were necessary to adopt the container

and would otherwise be captured by the adoptcontit variable set out in equation (2). The ijk fixed

effects that are removed when we first difference the estimating equation allow for differences in the

level of trade between country-pairs for each product k. To capture differences in the growth rate

between country-pairs, in Table 3 below, where we consider a number of robustness issues of our

main empirical findings, we add ij effects to the estimating equation.

3.3 Empirical findings

Although our entire data set covers a total of 157 countries, we initially report results for a

sub-set of 22 industrialized countries, which we denote as North-North trade, in Table 2.38 Because

we expect a quicker and deeper penetration of container technology in industrialized countries (recall

Figure 2), we initially apply equation (5) on an empirical domain where our technology variables are

expected to capture relatively similar transformations of the transportation sector. In addition,

35 As we will discuss in 3.3, we also considered an interaction term between the container variable and the regional trade

agreement variable.

36 See Bergstrand and Egger (2011) and Feenstra (2004, chapter 5) for good surveys of the gravity literature.

37 Since the country-time fixed effects preclude the inclusion of time varying ‘economic mass’ variables like GDP and

GDP per capita, we refrain from calling our specification a gravity equation.

38 Our selection criterion was OECD membership as of 1990. It is probably of no coincidence that these countries were

also almost all early adopters of the container and indeed that could be used as an alternative label for this group. The

countries are identified in Appendix Table 3.

21

restricting ourselves to North-North trade yields a full panel of country pairs. Afterwards, we extend

the empirical domain to the total sample of 157 countries, labeled as ‘World Trade’ (columns 3 and

4).

Our dependent variable is the change in the log of exports and imports between a country pair

i and j in a 4-digit product line k. We only considered observations with trade occurring in at least

one direction. The rationale behind this is that bilateral containerization should affect total bilateral

trade rather than a specific direction. Because not all country pairs trade the same set of products in

all years even in North-North trade, for which the panel is balanced on the ‘country-pair dimension’,

the panel is ’naturally’ imbalanced in the product dimension. However, we restricted our sample by

including only products which appear during the whole time period for at least one country pair.

Our complete set of panel estimates of equation (5), including the lagged variables, is

reported in columns (1) through (4) in Table 2. Depending on the specification, the number of

products in our regressions varies between 485 and 504 with the corresponding total number of

observations (or bilateral trade flows) varying between 227,501 (in regressions 1 and 2) and

1,169,501 (in regressions 3 and 4). All reported regressions include time-varying country fixed

effects for each origin country, it, each destination country, jt, and time-varying fixed effects for each

product, kt. Because our explanatory variables are more aggregated than our dependent variable, we

also cluster by country-pair-year ijt. Our key explanatory variables are the contemporaneous and

lagged effect of our binary container variable, Δcontij,t, and the contemporaneous and lagged effect of

this variable interacted with the product containerization variable, Δcontijt·prodk. Overall, the results

in Table 2 suggest that containerization is associated with statistically significant and economically

large changes to the growth of bilateral trade irrespective of whether we base our measure of

containerization on the timing of adoption or containerizability and irrespective of whether we

restrict the analysis to North-North or World-Trade. The coefficients of the lagged effects of Δcontij,t

reveal that containerization had strong and persistent effects in some cases even 15 years after its

bilateral adoption.

A key objective of our paper is to examine the causal effects of containerization on

international trade using non-containerized bilateral pairs as a counterfactual. One requirement for

such an interpretation is for the growth rate of trade to have been statistically similar for container

and non-containerized pairs before the adoption of the container took place. Following the advice of

Wooldridge (2002, p. 285), we add a future change of containerization, denoted by Δcontij,t+1, in our

regression equation (5). The size and statistical significance of Δcontij,t+1 can be viewed as a

falsification test for whether the container variable captures the effect of the introduction of this new

transportation technology rather than any trend to bilateral trade growth that was also present prior to

22

the adoption of containerization. If the effect captured by the container dummy were simply related

to growth trends already present in trade between that country-pair, we would expect the coefficient

on years prior to the adoption of the container to be as large and significant as the coefficient on our

variable of interest. Regression (2) in Table 2 also includes a pre-treatment variable on

containerization and the other bilateral covariates rta and GATT for North-North trade and regression

(4) performs the same exercise for the entire sample.

Regression (2) reveals that the coefficients of Δcontij,t+1 is statistically insignificant and also

very small compared to the contemporaneous and lagged variables, suggesting that the growth of

trade between non-containerized countries was statistically similar to that for containerized countries

prior to their containerization. Because the parallel trends assumption appears to be satisfied, the

estimates can be interpreted as capturing a causal effect of containerization on North-North trade, if

there are no other confounding factors that explain the decision to adopt the container and the growth

in trade for adopting-pairs. We note that this result is not robust to the extensions we describe in the

next section of the paper (Table 3). Hence, a causal interpretation is conditional on the specification.

Pre-adoption differences in trade flows are present in Table 2 when we extend the sample to

cover World Trade. In column (4) the coefficients on Δcontij,t+1 are statistically significant and also

relatively large. This implies that if one includes less developed economies (or late adopters), the

container coefficients are only suggestive of a high correlation as containerization was also adopted

in response to increased growth in trade. Overall, these results are consistent with the historical

narrative.

A somewhat different conclusion is reached for the containerizability interaction. The

coefficient on Δcontijt+1·prodk is statistically insignificant, indicating no pre-adoption difference in

the growth rate of trade between products that were containerizable compared to the counterfactual

of non-containerizable products. As a reminder this definition of containerizability uses the original

1968 German Engineers classification. This occurs for North-North trade in regression (2) and World

trade in regression (4). To conclude from these tables as a whole, an identification strategy that

exploits the product dimension of containerization appears a more promising route to capturing

causal relationships.

Let’s take a closer look at the estimates in our benchmark specification (2) for North-North

trade. The estimated coefficient on the cont variable in the upper panel suggests that the concurrent

effect of containerization was associated with a rise to logged bilateral trade flows by 0.987 log

points over 5-years compared to bilateral flows where both countries had not yet adopted the

container technology. To put this in a broader perspective, the mean of the dependent variable in the

dataset is equal to 0.570 log points with a standard deviation of 1.275. It would appear that

23

containerization is capable of explaining a large part of the differences in growth rates of trade over

time.

The coefficients on the lag variables reveal that over the next 5-year periods the effect of

containerization was 0.842 and 0.766 log-points respectively. The contemporaneous and lagged

effects of containerization sum to 2.595 (natural) log points, which equates to an annual growth rate

of trade of 17.3% for a 15-year period. Using the mean (median) value for product-bilateral trade

flows in our dataset of $2.5 million ($0.078 million) and our estimate of the cumulative ATE of

1,240% (=e2.595-1), this would imply that trade flows increased to $33.5 million ($1 million) because

of containerization after 15-years.

The estimated coefficient on the interaction variables cont·prod in column (2) is noticeably

smaller than for those based on the adoption of the container. From the discussion in Section 2.2 this

was not unexpected and indicates that many of the changes that occurred during adoption of the

container were not specific to containerizable products. This regression provides the smallest

estimate of the trade effect derived from the introduction of the container within the paper. In column

(2) the concurrent effect of containerization was to increase bilateral trade by 0.161 log-points for

containerizable products relative to non-containerizable products (equivalent to an annual growth

rate of 3.2% for 5-years, or a cumulative ATE of 17.4%). The lagged interaction effects are not

statistically significant, indicating that the growth rates of trade flows in containerizable and non-

containerizable products were statistically similar beyond this 5-year period. This result suggests that

containerization gave a permanent boost to the level of bilateral trade in containerizable products

through a short-lived increase to the growth rate of trade.

How do the effects of containerization compare to the time-varying trade policy variables that

are also included in the regression? As mentioned above, we included two sets of policy variables.

The rta dummy indicates whether a country pair ij belonged either to the same regional free trade

block or had a free trade agreement in a specific year. The GATT variable switches to 1 if both

countries i and j are members of the GATT, the precursor of the WTO, at time t. The inclusion of

lagged effects permits us also to investigate the dynamic effects of these variables.

Overall, we find that the estimated effects of containerization are generally much bigger than

the estimated effects of the trade policy variables in all specifications. This occurs even if we allow

for interaction effects between the container technology and the various trade policy variables, the

coefficients for which are insignificantly different from zero.39 The concurrent and the first two lags

of the rta variable have the expected positive sign and are highly significant. The concurrent effect of

39 For this reason we choose not to report the results from this regression. The results are available from the authors on

request.

24

a free trade agreement is to raise trade by an average of 0.313 log points (an increase to the annual

growth rate of 6.2%), which is about a third of the concurrent effect of containerization when

measures by the timing of adoption. The coefficients on the lags of the rta variable reveal that over

the next 5-year periods the effect was to increase the log of bilateral-product trade by 11% (=e0.104-1)

and 11% (=e0.103-1) respectively. The cumulative ATE of a free-trade agreement amount then to an

increase of about 68% at the end of 15-years compared to pre-treatment values. It is reassuring that

our cumulative ATE estimates of the rta variable is similar in magnitude to the estimates reported by

Baier and Bergstrand (2007, p.91), who consider a panel data on aggregate trade flows.

The concurrent and three lag effects of the GATT are all statistically significant and lie in

economic magnitude between the container and the rta variables. The concurrent effect of bilateral

GATT membership is to raise trade by an average of 52% (=e0.420-1) over 5-years, which is 15%

higher than the corresponding coefficient on the rta variable. The coefficients on the lags of the

GATT variable reveal relatively persistent long-term effects, 10+ years following bilateral

membership. Over the next 5-year periods the effect was 53% (=e0.428-1) and 26% (=e0.230-1)

respectively. The cumulative ATE of GATT membership for bilateral trade is then estimated to

amount to 194%, considerably higher than the average effect on free trade agreements, but less than

a sixth of the accumulated effect of containerization.

Regarding the presence of pre-treatment effects, column (2) shows that the coefficient of the

pre-treatment variable Δrtaij,t+1 is both statistically significant and relatively large, 0.128 log points

i.e. a 5-year growth rate of 14% (=e0.128-1). This suggests the presence of anticipation effects of

regional trade agreements in our sample. Column (4) reveals no anticipation effects for the world

sample. The estimated coefficient of the pre-treatment variable ΔGATTij,t+1 is statistically

insignificant for the North-North sample but not for the entire sample.

As mentioned before, the statistical significance of the estimate of the pre-treatment variable

Δcontij,t+1 in regression (4) is only suggestive of a high correlation between containerization and

international trade for the entire sample of 157 countries. However, a comparison of the estimated

concurrent and lagged container coefficients for the North-North and World sample reveal

interesting differences from including trade with developing economies. Although the concurrent

effect of containerization rises from 0.987 (a 5-year growth rate of 168%) to 1.310 (a 5-year growth

rate of 271%), the effect is much lower in the later periods. The cumulative ATE for World trade is

thus lower than estimated for North-North trade at 900%. This provides quantitative evidence for the

notion that the effects of containerization was more short-lived for developing economies’ trade as

they lacked the manufacturing base and/or the internal transportation infrastructure in order to take as

much advantage of this new technology as the industrial world. The interaction variable cont·prod in

25

regression (4) in contrast, suggests no contemporaneous difference between the growth of trade in

containerizable versus non-containerizable products amongst country-pairs that containerized, but

significant lagged effects. According to the estimates reported in column (4) trade increased by 0.079

log points 5-10 years after containerization and a further 0.053 log points in the following 5-years.

Together this indicates an cumulative ATE of 14% for World Trade because of containerization, an

effect that is a little smaller than for North-North trade in regression (2) and that takes a longer time

period to accrue.

3.4 Robustness

In Table 3 we consider the robustness of our main results. We begin by examining the

robustness to changes in the measure of containerizability. The first four regressions in Table 3 differ

from the regressions in Table 2 by adjusting the product containerizability measure with reported

mode of shipment information provided by the 2003 US Census as described in section 3.1.

Columns (1) and (3) report the results from adjustments to the non-containerizable component of the

containerizability variable, while columns (2) and (4) adjust both the containerizable and non-

containerizable components. As a reminder, we remove from the 1968 non-containerizable

classification products that are transported in containers in 2003 in regressions (1) and (3). In

addition, in regressions (2) and (4) we also remove from the 1968 containerizable classification

products that are not typically transported in containers in 2003.

For the North-North sample, this reduces the total product sample from 485 (Table 2 column

(2)) to 387 in column (1) and 316 in column (2). For World Trade, the product sample drops from

504 (Table 2 column (4)) to 399 in column (3) to 328 in column (4).

An obvious consequence of adjusting the set of containerizable and non-containerizable

products to include information on revealed container usage is to increase the estimated coefficients

of cont·prod from 0.161 (Table 2 column (2)) to 0.425 in regression (1) and 0.413 in regression (2).

This is equivalent to an annual growth rate of trade of 8.5% and 8.3% respectively, compared to an

estimate of just 3.2% in Table 2. A significant lagged effect to trade in containerizable products is

now also found in regression (1), indicating that the effects of the container technology on trade were

present 5-10 years after the adoption of the container. As a consequence the total effects of the

container in this regression are higher than previous estimates and indicate that trade was 85% higher

after 10-years because of the container.

The effects of containerization on changes to World trade also appear stronger when we

adjust the containerizability variable. In regression (3) trade in containerizable products is estimated

26

to have risen by 0.201 log points 5-10 years following the adoption of the container. The effect is

mildly higher in regression (4) at 0.211, which equates to a 23.5% increase over these 10-years.

Our next robustness test allows for differences in the rate of growth of trade between country-

pairs through the inclusion of ij fixed effects. We report these regressions in column (5) in Table 3.

Compared to regression (2), the inclusion of ij effects acts to increase the magnitude of the estimated

effects of containerization, both when measured by the timing of adoption of the interaction with

containerizability. For the various lags of Δcontij,t the accumulated effect of the container is an

estimated 3.1 log points over 15-years, equivalent to an average annual of 21% compared to trade

between country-pairs that had not adopted the container. Trade in containerizable products also

grew more quickly than non-containerized trade for 10-years after the adoption of the container in

this regression, by 0.30 log points (equivalent to an annual growth rate of 3% over the 10-years).

Again the effects of the container would appear larger when measured by differences changes in

trade between countries across time, compared to across products.

As a final exercise, in columns (6) and (7) we include only observations where trade flows in

any given country-pair-product combinations are positive in at least 4 of our 7 time periods.40 Here

we examine whether our results are driven by products being traded between countries on an

infrequent basis. The results for the contemporaneous and lagged effects appear robust to this

restriction on the data, both for North-North in column (6) and World trade in column (7). However,

column (6) reveals a violation of the parallel trends assumption, as revealed by a statistically

significant estimate on the coefficient of Δcontij,t+1. Now we find consistent evidence of a difference

in the pre-treatment growth rates of trade between adopters of the container and non-adopters.

Unexpectedly, these effects are negative in column (6), suggesting that trade growth was slower

between pairs of countries which later adopted the container. The results for World trade in column

(7), display little change compared to Table 2. However, we continue to find no evidence of a pre-

treatment difference in the growth of trade for the interaction with the product containerizability

variable in columns (6) and (7).

4. Conclusion

International trade is a key dimension of globalization and globalization plays a central role

for economic development and economic growth. But what are the drivers of international trade and

globalization? Business experts and historians who have studied or made a living from the shipment

40 A formal analysis of the impact of containerization on the extensive margin of trade is beyond the scope of this paper

and is left for future work.

27

of goods across international borders have long conjectured that “the shipping container made the

world smaller and the world economy bigger” (Levinson’s (2006) subtitle). In his recent world

history of technology, Daniel Headrick (2009, p.146) discusses containerization as the major 20th

century technological change that “…has propelled the globalization of the world economy”. To the

best of our knowledge, we are the first attempt to examine these claims in a rigorous manner. Our

identification strategy provides evidence that containerization is highly correlated with globalization

over the 20th century. We find that the growth in trade between countries occurs with the option of

the container, although our results caution against concluding that these effects are causal. Such a

claim is more credible when we exploit the fact that not all products that were traded internationally

were suitable to be placed inside of the container. However, the magnitude of the estimated effects

are much small than those based on the timing of adoption. Given the magnitude of the coefficients

for the container using the timing of adoption, exploration of cases which satisfy the conditions for

causal inference are an appealing avenue for future research. The success of this search might be

improved by exploiting the product dimension of containerization.

The findings in this paper should be of interest beyond academia. In assessing its support for

trade, the World Bank has recognized the importance of trade barriers beyond tariffs (World Bank

Independent Evaluation Group, 2006). In fact, a significant portion of the World Bank’s lending

budget is allocated to transportation infrastructure projects41. However quantitative studies on the

impacts of transportation infrastructure on the magnitude and patterns of international specialization

are still in its infancy. The estimates in our study suggest that the effects of the adoption of port

containerization on trade involving developing economies were relatively small compared to trade

among industrialized countries characterized by better domestic infrastructures. Our identification

strategy allowed us to draw inferences for what might be called ‘the early period of the container

age: 1966-1990’. We leave it for future research to assess the effects of containerization for the post

1990 period.

Acknowledgements

We acknowledge financial support from the Leverhulme Trust Grant F/00 114/AM. El-Sahli also

gratefully acknowledges financial support from Forte and Norface. Earlier versions of this paper

were presented at a 2011 CES-ifo Economic Studies Conference, the 2011 ETSG meetings in

Copenhagen, a CES-ifo-GEP 2011 joint conference in Munich, the 2013 CAGE trade workshop at

Warwick University, the 2013 CEPR-ERWIT conference in Rotterdam, the 2014 Washington area

international trade symposium and at seminars at the World Bank, Aarhus University, American

University, University of Bordeaux, City University Hong Kong, FIW Vienna, Florida International

University, Lund University, University of Nottingham-UK, University of Nottingham-Ningbo,

41 A large portion of the 2010 African Development Report is devoted to port capacity and containerization.

28

Shanghai University of Finance and Economics, Strathclyde University, Syracuse University and

Zhejiang University. We benefitted from discussions with Jeff Bergstrand, Sourafel Girma,

Wolfgang Keller, Thierry Maier, Chris Milner and Richard Upward. We are also very grateful to the

co-editor Daniel Trefler to four anonymous referees.

29

References

African Development Bank Group. 2010. African Development Report: Port, logistics, and trade in

Africa. Oxford University Press.

Annual digest of port statistics,1968-1979. London: National Ports Council.

Baier, Scott and Jeffrey Bergstrand, 2001. The growth of world trade: tariffs, transport costs, and

income similarity. Journal of International Economics 53(1), 1-27.

Baier, Scott and Jeffrey Bergstrand, 2007. Do free trade agreements actually increase members'

international trade? Journal of International Economics 71, 72-95.

Bergstrand, Jeffrey and Peter Egger, 2011. Gravity equations and economic frictions in the world

economy. In Palgrave Handbook of International Trade, edited by Daniel Bernhofen, Rod

Falvey, David Greenaway and Udo Kreickemeier. Palgrave MacMillan.

Blonigen, Bruce and Wesley Wilson, 2008. Port efficiency and trade flows. Review of International

Economics 16(1), 21-36.

Clark, Ximena, Dollar David and Alejandro Micco, 2004. Port efficiency, maritime transport costs

and bilateral trade. Journal of Development Economics 75(2), 417-450.

Containerisation International, 1970-1992. Various issues. London: National Magazine Co. Ltd.

Donaldson, Dave, 2012. Railroads of the Raj: Estimating the impact of transportation infrastructure.

MIT mimeo.

Donovan and J. Bonney, 2006. The Box That Changed the World: Fifty Years of Container

Shipping - An Illustrated History. Commonwealth Business Media Inc. .

Eyre, John, L. 1964. Shipping Containers in the Americas. In Pan American Union: Recent

Developments in the Use and Handling of Unitized Cargoes, Washington, D.C: 38-42.

Feenstra, Robert, 2004. Advanced International Trade : Theory and Evidence. New Jersey,

University Presses of California, Columbia and Princeton.

Feenstra, Robert, Robert E. Lipsey, Haiyan Deng, Alyson C. Ma and Hengyong Mo, 2005. World

Trade Flows: 1962-2000. NBER Working Paper No. 11040.

Fogel, Robert, 1964. Railways and American Growth, Johns Hopkins Press.

Headrick, Daniel, 2009. Technology: A World History. Oxford University Press.

Hummels, David, 2007. Transportation costs and international trade in the second era of

globalization, Journal of Economic Perspectives 21, 3:131-154.

Hurd, J., 1975. Railways and the expansion of markets in India, 1861-1921. Explorations in

Economic History 12, 263-288.

30

Krugman, Paul, 1995. Growing world trade: causes and consequences. Brookings Papers on

Economic Activity, 1995(1):327-377.

Krugman, Paul, 2009. Citigroup Foundation Special Lecture, Festschrift paper in Honor of Alan V.

Deardorff, University of Michigan IPC working paper 91.

Kuby Michael and Neil Reid, 1992. Technological change and the concentration of the U.S. general

cargo port system: 1970-88. Economic Geography 68, 272-289.

Levinson, Marc. 2006. The Box: How the Shipping Container Made the World Smaller and the

World Economy Bigger. Princeton University Press.

McKinsey Co, 1967. Containerization: the key to low-cost transport. A report for the British

Transport Docks Board.

McKinsey Co., 1972. Containerization: A 5-year balance sheet.

Mokyrl, Joe, 1992. The Lever of Riches: Technological Creativity and Economic Progress. Oxford

University Press.

National Ports Council, 1968-1979. Annual digest of port statistics. London: National Ports Council.

Notteboom, Theo and Jean-Paul Rodrigue, 2008. Containerisation, box logistics and global supply

chains: the integration of ports and liner shipping networks. Maritime Economics & Logistics

10, 152-174.

Raine, George, 2006. “A sea change in shipping: 50 years ago, container ships altered the world”,

San Francisco Chronicle, February 5.

Rua, Gisela, 2014. Diffusion of containerization. Finance and Economics Discussion Series Staff

working paper 2014-88. Federal Reserve Board, Washington, DC.

Stopford, M. 2009. Maritime Economics , 3rd edition, Routledge.

Wooldridge, Jeffrey, 2002. Econometric Analysis of Cross-Section and Panel Data. MIT Press,

Cambridge, MA.

World Bank Independent Evaluation Group. 2006. Assessing Word Bank support for trade, 1987-

2004. Washington, D.C: The World Bank.

Yi, Kei-Mu, 2003. Can vertical specialization explain the growth in world trade? Journal of

Political Economy 111(1), 52-102.

31

Table 2: Benchmark estimates

Standard errors in parentheses

* p<0.05, ** p<0.01, *** p<0.001

𝑎!Total ATE is the sum of statistically significant container estimates.

Dependent Variable is the (1st differenced) log of trade between countries i (origin country) and j (destination) in product k; t denotes time.

Regressions (1) to (4) use the German Engineers classification of 1968 to classify products as containerizable or non-containerizable.

!

North-North

World trade

Regression No.

(1)

(2)

(3)

(4)

∆!𝒄𝒐𝒏𝒕𝒊𝒋,𝒕

0.990***

0.987***

1.238***

1.310***

(0.154)

(0.159)

(0.037)

(0.036)

∆!𝒄𝒐𝒏𝒕𝒊𝒋,𝒕!𝟏

0.845***

0.842***

0.634***

0.693***

(0.098)

(0.099)

(0.033)

∆!𝒄𝒐𝒏𝒕𝒊𝒋,𝒕!𝟐

0.768***

0.766***

0.264***

0.300***!

(0.083)

(0.084)

(0.027)

(0.027)!

∆!𝐜𝐨𝐧𝐭𝒊𝒋,𝒕!𝟏

-0.060

0.302***!

(0.135)

(0.021)!

∆!𝐜𝐨𝐧𝐭𝒊𝒋,𝒕∙𝐩𝐫𝐨𝐝𝒌

0.153**

0.161**

0.006

0.007

(0.062)

(0.079)

(0.014)

(0.015)

∆𝐜𝐨𝐧𝐭𝒊𝒋,𝒕!𝟏∙𝐩𝐫𝐨𝐝𝒌

0.081

0.085

0.078***

0.079***

(0.057)

(0.061)

(0.014)

∆𝐜𝐨𝐧𝐭𝒊𝒋,𝒕!𝟐∙𝐩𝐫𝐨𝐝𝒌

-0.045

-0.043

0.053***

(0.045)

(0.047)

(0.013)

∆!𝐜𝐨𝐧𝐭𝒊𝒋,𝒕!𝟏∙𝐩𝐫𝐨𝐝𝒌

0.018

-0.001

(0.097)

(0.015)

∆!𝐫𝐭𝐚𝒊𝒋,𝒕

0.306***

0.313***

0.322***

0.334***!

(0.030)

(0.034)

(0.034)!

∆!𝐫𝐭𝐚𝒊𝒋,𝒕!𝟏

0.098***

0.104***

0.136***

0.143***!

(0.033)

(0.030)

(0.031)!

∆!𝐫𝐭𝐚𝒊𝒋,𝒕!𝟐

0.103***

0.040

0.044!

(0.030)

(0.030)!

∆!𝐫𝐭𝐚𝒊𝒋,𝒕!𝟏

0.128***

0.052!

(0.036)

(0.034)!

∆!𝐆𝐀𝐓𝐓

𝒊𝒋,𝒕

0.415***

0.420***

0.337***

0.326***!

(0.118)

(0.031)

(0.031)!

∆!𝐆𝐀𝐓𝐓

𝒊𝒋,𝒕!𝟏

0.422***

0.428***

0.117***

0.112***!

(0.095)

(0.026)

(0.025)!

∆!𝐆𝐀𝐓𝐓

𝒊𝒋,𝒕!𝟐

0.223***

0.230***

-0.013

-0.014!

(0.073)

(0.024)

(0.024)!

∆!𝐆𝐀𝐓𝐓

𝒊𝒋,𝒕!𝟏

0.072

0.059*!

(0.190)

(0.029)!

Obs!

227501

1169501

Countries!

22

157

157!

No. Products

485

504

𝑅!

0.147

0.148

0.102

0.103!

FE

it,jt,kt

it,jt,kt!

Clustering

ijt

32

Table 3: Robustness

North-North

World trade

North-North

World

trade

Using German and US Census classification of product

containerizability

ij FE

Obs count

4-7

Obs count

4-7

Regression No.

(1)

(2)

(3)

(4)

(5)

(6)

(7)

∆!𝒄𝒐𝒏𝒕𝒊𝒋,𝒕

0.802***

0.865***

1.298***

1.248***

1.232***

1.093***

1.176***

(0.210)

(0.216)

(0.044)

(0.1451)

(0.187)

(0.044)

∆!𝒄𝒐𝒏𝒕𝒊𝒋,𝒕!𝟏

0.765***

0.756***

0.660***

0.640***

1.057***

0.780***

0.777***

(0.128)

(0.129)

(0.040)

(0.0747)

(0.097)

(0.036)

∆!𝒄𝒐𝒏𝒕𝒊𝒋,𝒕!𝟐

0.708***

0.706***

0.247***!

0.235***!

0.839***

0.736***

0.406***

(0.096)

(0.097)

(0.034)!

(0.035)!

(0.0588)

(0.083)

(0.028)

∆!𝐜𝐨𝐧𝐭𝒊𝒋,𝒕!𝟏

-0.286

-0.368*

0.265***!

0.255***!

-0.087

-0.463*

0.260***

(0.182)

(0.032)!

(0.031)

(0.1367)

(0.251)

(0.028)

∆!𝐜𝐨𝐧𝐭𝒊𝒋,𝒕∙𝐩𝐫𝐨𝐝𝒌

0.425***

0.413**

-0.004

0.008

0.186**

0.184**

0.023

(0.153)

(0.155)

(0.028)

(0.027)

(0.0789)

(0.084)

(0.016)

∆𝐜𝐨𝐧𝐭𝒊𝒋,𝒕!𝟏∙𝐩𝐫𝐨𝐝𝒌

0.189*

0.160

0.109***

0.112***

0.113*

0.134**

0.093***

(0.109)

(0.111)

(0.026)

(0.0600)

(0.062)

(0.015)

∆𝐜𝐨𝐧𝐭𝒊𝒋,𝒕!𝟐∙𝐩𝐫𝐨𝐝𝒌

0.025

0.004

0.092***

0.099***

-0.020

-0.029

0.055***

(0.080)

(0.082)

(0.025)

(0.024)

(0.0455)

(0.047)

(0.013)

∆!𝐜𝐨𝐧𝐭𝒊𝒋,𝒕!𝟏∙𝐩𝐫𝐨𝐝𝒌

0.314*

0.295

0.033

0.032

0.025

-0.002

0.019

(0.161)

(0.029)

(0.027)

(0.0991)

(0.099)

(0.017)

∆!𝐫𝐭𝐚𝒊𝒋,𝒕

0.320***

0.327***

0.331***!

0.395***

0.311***

0.312***

(0.031)

(0.034)!

(0.0323)

(0.030)

(0.032)

∆!𝐫𝐭𝐚𝒊𝒋,𝒕!𝟏

0.107**

0.104**

0.140***!

0.131***!

0.184***

0.104***

0.120***

(0.035)

(0.032)!

(0.033)!

(0.0361)

(0.034)

(0.031)

∆!𝐫𝐭𝐚𝒊𝒋,𝒕!𝟐

0.107***

0.111***

0.049!

0.051!

0.171***

0.100***

0.040

(0.031)

(0.032)

(0.031)!

(0.032)!

(0.0398)

(0.030)

(0.029)

∆!𝐫𝐭𝐚𝒊𝒋,𝒕!𝟏

0.122***

0.133***

0.051!

0.076**!

0.212***

0.130***

0.034

(0.037)

(0.038)

(0.035)!

(0.037)!

(0.0343)

(0.038)

(0.032)

∆!𝐆𝐀𝐓𝐓

𝒊𝒋,𝒕

0.372***

0.279*

0.304***!

0.294***!

0.267**

0.733***

0.352***

(0.131)

(0.136)

(0.032)!

(0.1141)

(0.148)

(0.038)

∆!𝐆𝐀𝐓𝐓

𝒊𝒋,𝒕!𝟏

0.380***

0.353**

0.111***!

0.122***!

0.291***

0.485***

0.156***

(0.107)

(0.114)

(0.026)!

(0.027)!

(0.0903)

(0.109)

(0.028)

∆!𝐆𝐀𝐓𝐓

𝒊𝒋,𝒕!𝟐

0.236***

0.235**

-0.017!

-0.012!

0.154**

0.224***

0.013

(0.078)

(0.080)

(0.024)!

(0.025)!

(0.0640)

(0.074)

(0.024)

∆!𝐆𝐀𝐓𝐓

𝒊𝒋,𝒕!𝟏

0.072

0.122

0.058!

0.047!

0.021

-0.143

0.078**

(0.216)

(0.226)

(0.031)!

(0.032)!

(0.1912)

(0.301)

(0.037)

Obs!

183887

153665

945776

821403

226241

229382

996452

Countries!

22

157!

22

157

No. Products

387

316

399

328

485

482

494

𝑅!

0.152

0.154

0.106!

0.109!

0.152

0.149

0.110

FE

it,jt,kt!

it,jt,kt,ij

it,jt,kt

Clustering

ijt

Standard errors in parentheses

* p<0.05, ** p<0.01, *** p<0.001

𝑎!Total ATE is the sum of statistically significant container estimates.

Dependent Variable is the (1st differenced) log of trade between countries i (origin country) and j (destination) in product k; t denotes time.

Regressions (5) and (7): Using German Engineers classification supplemented with US Census trade data of 2003. From the latter, we measure containerizability of a

product k as containerized trade(k)/total trade(k). We then drop from the non-containerizable groups those products that had containerizability measure in 2003 larger than

the bottom quartile of containerizability (i.e. in the top three quartiles).

Regressions (6) and (8): In addition to the above for regressions (5) and (7), we also drop from the treated (containerizable group) those products that become non-

containerizable (products that have a containerizability measure in the bottom quartile).

Regressions (9) to (11) use the German Engineers classification of 1968 to classify products as containerizable or non-containerizable.

33

Appendix

Appendix Table 1: Variables and Data Sources

Variables Data Sources

Trade Flows

Feenstra et al. (2005)

Container variables

Containerisation International Yearbook

(several years)

Policy variables

CEPII, Paris

US trade flows 2003

US Census (https://usatrade.census.gov)

Appendix Table 2: Correlations between variables

North-North

contij

rta

GATT

contij

1

rta

0.260

1

GATT

0.331

0.119

1

Entire Sample

contij

rta

GATT

contij

1

rta

0.199

1

GATT

0.278

0.229

1

34

Appendix Table 3: Countries in the sample

Panel A: Countries that containerize by port or rail 1966-1983 (122 countries)

1966

West Germany(P)*

Netherlands(P)*

UK(P)(R)*

USA(P)*

India (R)

1968

Australia(P)*

Austria(R)*

Belgium(P)*

Canada(P)*

Denmark(P)*

East Germany(R)

France(P)*

Hungary(R)

Ireland(R)*

Italy(P)*

Spain(R)*

Sweden(R)*

Switzerland(R)*

Taiwan(P)

1969

Finland(P)*

Yugoslavia(R)

Japan(P)*

Norway(R)*

Portugal(P)*

1970

Hong Kong(P)

USSR(R)

Greece(P)*

Israel(P)

Romania(R)

Singapore(P)

1971

Cote D’Ivoire(P)

New Zealand(P)*

Philippines(P)

Poland(P)

Trinidad(P)

1972

Bulgaria(R)

Czechoslovakia(R)

1973

Bahamas(P)

Brazil(P)

Iceland(P)*

Jamaica(P)

Malaysia(P)

1974

Cameroon(P)

Chile(P)

Colombia(R)

Nigeria(P)

Panama(R)

South Africa(P)

1975

Barbados(P)

Honduras(P)

Indonesia(P)

Korea Rep(P)

Peru(P)

Thailand(P)

1976

Argentina(P)

Benin(P)

Kenya(P)

Mexico(P)

N. Caledonia(P)

Saudi Arabia(P)

UAE(P)

1977

Bahrain(P)

Cyprus(P)

Ghana(P)

Iran(P)

Jordan(P)

Kuwait(P)

Lebanon(P)

Morocco(P)

1978

Ecuador(P)

Egypt(P)

Gibraltar(P)

Haiti(P)

Iraq(P)

Mozambique(P)

Oman(P)

Papua N.

Guinea(P)

Samoa(P)

Sierra Leone(P)

St. Kitts Nevis(P)

Tanzania(P)

1979

Algeria(P)

Angola(P)

China(P)

Congo(P)

Djibouti(P)

El Salvador(P)

Mauritius(P)

Neth.Antilles(P)

Nicaragua(P)

Pakistan(P)

Qatar(P)

Sri Lanka(P)

Syria(P)

1980

Guatemala(P)

Liberia(P)

Libya(P)

Madagascar(P)

Sudan(P)

Uruguay(P)

1981

Brunei/Bhutan(P)

Bangladesh(P)

Belize(P)

Costa Rica(P)

Dem.Rep.Congo(P)

Dominican Rep(P)

Fiji(P)

Guadeloupe(P)

Seychelles(P)

Togo(P)

Tunisia(P)

Turkey(P)

Venezuela(P)

1982

Gambia(P)

Kiribati(P)

Mauritania(P)

St. Helena(P)

1983

Bermuda(P)

Ethiopia(P)

Guinea(P)

Malta(P)

Myanmar(P)

(P) denotes that the country containerized by port first.

(R) denotes that the country containerized by rail first.

(*) denotes that the country is in the North-North sample.

Panel B: Countries that do not containerize by port or rail 1966-1983 (35 countries)

Afghanistan

Chad

Greenland

Mongolia

Senegal

Albania

Cuba

GuineaBissau

Nepal

Somalia

Bolivia

Eq. Guinea

Guyana

Niger

Suriname

Burkina Faso

Falkland Islands

Laos

North Korea

Uganda

Burundi

French Guiana

Macao

Paraguay

Viet Nam

Cambodia

French Overseas

Malawi

Rwanda

Zambia

Cen. African Rep

Gabon

Mali

St. Pierre Miquelon

Zimbabwe

35

Appendix Table 4: List of industries42

Panel A

Class A: Containerizable products (4-digit SITC in 1968, 712 products)

Code

Good Description (number of underlying 4-digit products)

035

Fish, dried, salted or in brine smoked fish (1)

037

Fish, crustaceans and molluscs, prepared or preserved (3)

042

Rice (3)

046

Meal and flour of wheat and flour of meslin (1)

047

Other cereal meals and flours (1)

048

Cereal preparations & preparations of flour of fruits or vegetables (6)

056

Vegetables, roots & tubers, prepared/preserved, n.e.s. (4)

058

Fruit, preserved, and fruit preparations (1)

061

Sugar and honey (6)

062

Sugar confectionery and other sugar preparations (1)

071

Coffee and coffee substitutes (3)

072

Cocoa (4)

073

Chocolate & other food preparations containing cocoa (1)

074

Tea and mate (3)

075

Spices (3)

081

Feedstuff for animals (not including unmilled cereals) (6)

091

Margarine and shortening (3)

098

Edible products and preparations n.e.s. (1)

111

Non alcoholic beverages, n.e.s. (1)

112

Alcoholic beverages (5)

121

Tobacco, unmanufactured; tobacco refuse (4)

122

Tobacco manufactured (4)

211

Hides and skins (except furskins), raw (7)

212

Furskins, raw (including astrakhan, caracul, etc.) (1)

222

Oil seeds and oleaginous fruit (excluding flours and meals) (7)

223

Oils seeds and oleaginous fruit, whole or broken (including flours and meals) (7)

23

Crude rubber (including synthetic and reclaimed) (5)

244

Cork, natural, raw & waste (including in blocks/sheets) (1)

25

Pulp and waste paper (7)

26

Textile fibres (except wool tops) and their wastes (30)

277

Natural abrasives, n.e.s (including industrial diamonds) (3)

291

Crude animal materials, n.e.s. (3)

411

Animal oils and fats (3)

423

Fixed vegetable oils, soft, crude, refined/purified (7)

424

Other fixed vegetable oils, fluid or solid, crude (7)

431

Animal & vegetable oils and fats, processed & waxes (5)

53

Dyeing, tanning and colouring materials (11)

54

Medicinal and pharmaceutical products (8)

55

Essential oils & perfume materials; toilet polishing and cleansing preparations (9)

58

Artificial resins, plastic materials, cellulose esters and ethers (29)

59

Chemical materials and products, n.e.s. (13)

61

Leather, leather manufactures, n.e.s. and dressed furskins (13)

62

Rubber manufactures, n.e.s. (13)

63

Cork and wood manufactures (excluding furniture) (13)

64

Paper, paperboard, articles of paper, paper-pulp/board (15)

65

Textile yarn, fabrics, made-up articles, related products (61)

664

Glass (10)

665

Glassware (5)

42 Here we list all industries classified by the Verband Deutscher Ingenieure (German Association of Engineers). In the regressions

that use this classification, we only consider class A products to be suitable for containers (containerizable) whereas we consider

products in classes B and C to be not suitable for containers (non-containerizable). Because of few trading years and missing values

our regressions are run on a smaller subset of products. For example, class A covers 712 product lines, but our regressions in Table 2

are run only on 485 product lines for North-North trade.

36

666

Pottery (1)

667

Pearls, precious & semi-prec.stones, unwork./worked (5)

673

Iron and steel bars, rods, angles, shapes & sections (5)

674

Universals, plates and sheets, of iron or steel (8)

675

Hoop & strip, of iron/steel, hot-rolled/cold-rolled (1)

677

Iron/steel wire, wheth/not coated, but not insulated (1)

678

Tubes, pipes and fittings, of iron or steel (6)

679

Iron & steel castings, forgings & stampings; rough (1)

681

Silver, platinum & oth.metals of the platinum group (3)

682

Copper (3)

683

Nickel (3)

684

Aluminium (3)

685

Lead (3)

686

Zinc (3)

687

Tin (3)

689

Miscell.non-ferrous base metals employ.in metallurgy (3)

692

Metal containers for storage and transport (3)

693

Wire products and fencing grills (4)

694

Nails, screws, nuts, bolts etc.of iron, steel, copper (1)

695

Tools for use in hand or in machines (5)

696

Cutlery (2)

697

Household equipment of base metal, n.e.s. (5)

699

Manufactures of base metal, n.e.s. (10)

71

Power generating machinery and equipment (26)

723

Civil engineering and contractors plant and parts (4)

724

Textile & leather machinery and parts (7)

725

Paper and pulp mill mach., mach for manuf.of paper (4)

726

Printing and bookbinding mach.and parts (6)

727

Food processing machines and parts (2)

728

Mach. & equipment specialized for particular ind. (5)

73

Metalworking machinery (11)

741

Heating & cooling equipment and parts (6)

742

Pumps for liquids, liq.elevators and parts (7)

743

Pumps & compressors, fans & blowers, centrifuges (8)

7449

Parts of the machinery of 744.2- (1)

745

Other non-electrical mach.tools, apparatus & parts (2)

749

Non-electric parts and accessories of machines (5)

75

Office machines & automatic data processing equipment (15)

76

Telecommunications & sound recording apparatus (16)

77

Electrical machinery, apparatus & appliances n.e.s. (31)

7840

Parts & accessories of 722--, 781--, 782--, 783—(1)

7929

Parts of heading 792--, excl.tyres, engines (1)

8

Miscellaneous manufactured articles (114)

!

Panel B

Class B: Products with limited containerizability (4-digit SITC in 1968, 126 products)

Code

Good Description (number of underlying 4-digit products)

01

Meat and meat preparations (15)

02

Dairy products and birds' eggs (9)

034

Fish, fresh (live or dead), chilled or frozen (5)

036

Crustaceans and molluscs, fresh, chilled, frozen etc. (1)

054

Vegetables, fresh, chilled, frozen/preserved; roots, tubers (7)

057

Fruit & nuts (not including oil nuts), fresh or dried (9)

248

Wood, simply worked, and railway sleepers of wood (4)

271

Fertilizers, crude (5)

37

287

Ores and concentrates of base metals, n.e.s. (9)

288

Non-ferrous base metal waste and scrap, n.e.s. (3)

292

Crude vegetable materials, n.e.s. (8)

51

Organic chemicals (29)

52

Inorganic chemicals (14)

671

Pig iron, spiegeleisen, sponge iron, iron or steel (4)

691

Structures & parts of struc.; iron, steel, aluminium (4)

Panel C

Class C: Non-containerizable products (4-digit SITC in 1968, 152 products)

Code

Good Description (number of underlying 4-digit products)

001

Live animals chiefly for food (7)

041

Wheat (including spelt) and meslin, unmilled (3)

043

Barley, unmilled (1)

044

Maize, unmilled (1)

045

Cereals, unmilled (no wheat, rice, barley or maize) (4)

245

Fuel wood (excluding wood waste) and wood charcoal (1)

246

Pulpwood (including chips and wood waste) (1)!

247

Other wood in the rough or roughly squared (4)

273

Stone, sand and gravel (5)

274

Sulphur and unroasted iron pyrites (3)

278

Other crude minerals (7)

281

Iron ore and concentrates (4)

282

Waste and scrap metal of iron or steel (2)

289

Ores & concentrates of precious metals; waste, scrap (1)

3

Mineral fuels, lubricants and related materials (25)

56

Fertilizers, manufactured (5)

57

Explosives and pyrotechnic products (5)

661

Lime, cement, and fabricated construction materials (5)

662

Clay construct.materials and refractory constr.mater (3)

663

Mineral manufactures, n.e.s (8)

672

Ingots and other primary forms, of iron or steel (5)

676

Rails and railway track construction material (1)

721

Agricultural machinery and parts (5)

722

Tractors fitted or not with power take-offs, etc. (3)

744

Mechanical handling equip.and parts (5)

781

Passenger motor cars, for transport of pass., goods (1)

782

Motor vehicles for transport of goods and materials (3)

783

Road motor vehicles, n.e.s. (3)

7841

Chassis fitted with engines for motor vehicles (1)

7842

Bodies for the motor vehicles of 722/781/782/783 (1)

785

Motorcycles, motor scooters, invalid carriages (4)

786

Trailers and other vehicles, not motorized (4)

791

Railway vehicles and associated equipment (7)

7921

Helicopters (1)

7922

Aircraft not exceeding an unladen weight 2000 kg (1)

38

7923

Aircraft not exceeding an unladen weight 15000 kg (1)

7924

Aircraft exceeding an unladen weight of 15000 kg (1)

7928

Aircraft, n.e.s.balloons, gliders etc and equipment (1)

793

Ships, boats and floating structures (5)

9

Commodities and transactions not elsewhere classified (7)