Transcript Railways and the Raj: The impact of colonial infrastructure

Railways and the Raj: The
Economic Impact of
Transportation Infrastructure
Dave Donaldson
([email protected])
Research Questions
• What is the effect on economic outcomes
of opening up to external (ie. international)
trade?
• What is the effect on economic outcomes
of enabling internal (ie. inter-regional)
trade?
• What are the economic gains from
improving transportation infrastructure?
• Why economic change underpins these
effects?
Motivation
• Our understanding of the effects of openness to
trade is still incomplete:
– External trade: usually all of country liberalises trade
at same time, so finding counterfactuals is difficult
– Internal trade: virtually unexplored, for lack of data
• Transportation infrastructure is a dominant
important policy issue in LDCs (eg WDR 1994),
yet evidence base is lacking
– very hard to evaluate, due to endogenous placement
This paper
• Collect new dataset on prices, wages, production
(agricultural), and trade at the district-level (N=~300) in
India, from 1870-1925
• Use features of colonial construction of railways (18501900) in India as a set of ‘natural experiments’ in
openness
– Military motive (responding to domestic and foreign aggression)
– Famine-prevention motive
• Study impact of railways on agricultural output
• Interpret this impact in context of a simple trade model
– Predicts specialisation in comparative advantage crops
• Use data on internal and external trade flows to examine
this mechanism
• Where data permits, examine other possible
mechanisms (capital and labour reallocations,
technological change)
Why Colonial India?
• This region and period of history offer a number
of institutional and methodological advantages:
– Railway system was dramatic shock (in most of India
at this time, road and river transport was poor/nonexistent)
– Railway line placement motives were non-economic
in many instances
– Availability of unique internal trade data
• Allows external trade to be studied using within-country
variation
• Allows internal trade to be studied
Related Literature
• Effect of openness, using natural experiment
approach:
– Bernhofen-Brown (JPE ‘03, AER ‘04) use Japan’s
1851 (forced) openness to test comparative
advantage mechanisms behind opening
– Michaels (2006) uses US Interstate highway
expansion to study effect of openness on skill
premium
• Quantifying the gains from railways:
– Fogel (1967) on USA: uses ‘social savings’ technique,
ignores endogenous placement
– Hurd (1998) on India: same method; finds large effect
(9% of GDP in 1900)
This presentation
– Background:
• Railways
• Economic environment
– Elements of a simple theoretical framework
for thinking about these issues
– Data
– Empirical method
• Identification strategy for estimating effect of
railways
• What economic mechanisms underpin this effect?
Background: Railways
• Principal public investment in colonial
India (over half of government spending)
• Mixtures of pure public and public-private
provision, but Indian Government always
determined route selection
• 95% of current lines built in 1853-1930
• 1870-1920 was highest growth period
1870
65 districts had railway
somewhere in district
1900
170 districts had railway
somewhere in district
1930
220 districts had railway
somewhere in district
Background: Economic
Environment (1)
• Structure of economy in 1870:
– Agriculture: 68% of GDP, (73% of labour)
– Small-scale manuf. and services: 26%, (26%)
– Large-scale manufacturing: 0.5%, (0.2%)
• Structure of economy in 1930:
– Agriculture: 59%, (75%)
– Small-scale manuf. and services: 34%, (23%)
– Large-scale manufacturing: 4%, (2%)
Background: Economic
Environment (2)
• Effect of railways on transport costs:
– Standard estimates suggest that the pure
freight costs of railways were 5-10 times lower
than on alternative method (bullock carts)
– However, this ignores other savings:
• Bullocks/roads seasonal (bullocks need
food/water, roads unpassable for
Data (1870-1930)
• Agricultural production (annual, ~300 districts/native
states):
–
–
–
–
Yields, by crop (~15 crops)
Land area allocations, by crop
Capital stocks (livestock, carts)
Irrigated areas, by crop
• Prices and wages (annual, ~200 districts/native states)
– Prices: by ~30 commodities
– Wages: by ~5 occupations (skilled and unskilled)
• Trade (annual, ~70 trade blocks):
– Internal trade: full block-to-block matrix of trade flows (but intrablock diagonals empty)
– External trade: trade by port, by foreign country
– All in physical units, by commodity (~100 goods), by mode of
transportation (rail, river, coast)
Limitations of the Data
• Agricultural Yields:
– Subject of much controversy among econ historians
– Created by multiplying normal yields (factual) by subjective
‘conditioning factor’
– But largely corroborated by quinquennial crop-cutting surveys
(and no obvious signs that this is not just classical ME)
• Trade data:
– External trade flows by block not collected
• have to make assumptions of constant port consumption, and no
port transformation
– Roads data very limited in coverage
– Lack of unit values may obscure quality-differentiation within
observed commodity classifications
The second stage
• Run regressions of form:
ydt   d   t  Rdt  X dt   dt
y = real agricultural output
d = district
t = year
R = shortest distance from (populationweighted geographic) centre of district to
railway
X= other controls
• Can then think of modifying how R is included,
to allow for heterogeneous treatment
– Distance to port (and which port)
– Distance to internal cities, or other markets
The first stage
•Run regressions of form:
Rdt  d  t  Zdt  X dt  dt
•Where Z is a variable that predicts R, but
has no direct effect on y
General IV set-up (1)
• Railways are lines designed to connect two points, A and
B
• For any points (A,B), and the observed railway between
them, can ask:
– What is the effect of the railway on A or B?
– What is the effect of the railway on intervening point C?
C
RCdt = d
d
A
B
RAdt ~=0
RBdt ~=0
General IV set-up (2)
• Challenge is to find A-B pairs, such that:
– (1) the decision to put a railway between A and B had
nothing to do with unobservable characteristics of C
– (2) there is nothing unobservably different about
locations C along the line from A-B
• It is very unlikely that A or B can be used in the
analysis, for fear that exclusion restriction
violated there
– So ideally want 2 or more IVs, with very different
types of A-B pairs
Instrumental Variables (Option 1):
Famine-prevention in 1880
• 1880 Famine Commission recommended a
number of railways to be built
• This was idiosyncratic feature of that
Commission: earlier and later Famine
Commissions did not recommend any railways
• Translation into instrument
– A: locations of abnormally low rainfall in 1877-78
– B: nearest point to A that is on an 1879 railway line
– Control for rainfall variation (at C) throughout period
Lines suggested in
1880 Famine
Commission report
Instrumental Variables (Option 2) –
Military Transportation
• Macpherson (1955) estimates that over half of
track placement decisions were militarily-driven
• British government was motivated by internal
control, and external border defence (esp.
Afghanistan border)
• Translation to IV:
– A: sites of suspected military action, not already on a
railway at time t
– B: nearest military cantonment (base) to A, or nearest
point on existing railways to A
What mechanisms drive the result?
• Obtain a 2SLS estimate ˆ, but what is driving this change?
– Specialisation?
– Specialisation according to comparative advantage?
– Capital accumulation:
• returns to capital higher?
• railways affected banks’ ability to monitor borrowers?
– Labour supplied to agriculture changes?
• Higher wage draws in labour from other sectors?
• Railways enable migration?
– Land used in agriculture increases?
• Extension of land cultivation margin (deforestation etc.)?
• More double-cropping?
– Technological progress?
• Returns to innovation higher (size of market larger)?
• Technology transfer on the railways?
Conclusion
• Have presented plans for future research
designed to help address important gaps
in our understanding of external and
internal openness
– What is the effect of openness?
– What is driving this effect?