session 4 c discussion of xianjia ye paper in session 4a 26 august
DESCRIPTION
Parallel Session IARIW 2014TRANSCRIPT
Shifts in Comparative Advantage and Industrial Structure
when Production is Internationally Fragmented
Author: Xianjia Ye
Discussant: D.S. Prasada Rao
The main objective of the paper is to provide an explanation as to why different countries have different paths and speeds of structural change. Objectives are to:
Empirically describe structural change paths by investigating relatedness of activities;
To develop a measure of relatedness between various activities; Empirically implement the measure using the World Input Output
Database (WIOD); Identify the patterns implicit in the matrix of bilateral relatedness
indices; and Testing whether potential paths for upgrading have predictive power
on the direction of structural change
Objectives
Measuring relative comparative advantage (RCA) of an activity
Use of exports versus value-added export of an activity x in country a
Measuring value-added export of activity x in country a using WIOD data
Concept of relatedness between two activities Asymmetric versus symmetric measures
Identifying structures in the bilateral relatedness indices Heat maps; graph theoretic approach; clustering
methods Testing role of activity relatedness in structural change
Probit models; alternative specifications; robustness
Route map for the paper
The paper makes use of WIOD as the main data source covering 40 countries. The paper focuses on 28 out of 35 industries in the WIOD (excluded are
public sector and service industries) Value added contributions are considered for labour at three different skill
levels Low-skilled workers – with lower-secondary or lower level of education Medium-skilled labour – secondary level education or with post-secondary
but non-tertiary education (vocational training) High-skilled workers – university graduates with tertiary or higher level of
education The paper distinguishes a total of 84 activities = 28 (industries) x 3 skill
levels Paper constructs an 84 x 84 matrix containing the relatedness index between
each pair of activities – construction shown below. Activities 1 to 28 – LOW skill; 29 to 56 – MEDIUM skill; 57 to 84 HIGH
skill (1, 29 and 57 refer to agriculture; 2, 30 and 58 refer to mining and quarrying, etc.)
Structure of Activity Space
Export values versus Value added exports
Retail Price in US: $299Apple Profit + Retail Margin: 50% total Price
Export
30G IPod Classic (2005)Factory Cost: $144.56
Import
Hard Disk: $73.39Screen: $23.27Decoder Chip:$ 8.36Control Chip: $ 4.94Battery: $ 2.8932MB RAM: $ 2.378MB RAM: $ 1.85Flash RAM: $ 0.84Other Parts: $22.79------------------------------Subtotal: 140.70
Registered Export of China: Looks Good iPods, in Electronics Industry, $144.56, to mainly Rich Countries.
Actual Activities in China: Low-skilled, Low value-added Assembly the parts together, and test whether the iPods work.
Actual Values Captured By China: Low $3.86, appr. 3% of the total factory cost
(Source: Dedrick, Kraemer and Linden 2008)
Start with a world input-output table with m countries and n industries . This is represented by the input-output matrix of order mn by mn.
A typical element is the value of intermediate goods from industry x of country I that used in the production of process of $1 output of industry y in country j.
Computation of value-added export of each actvity
( , )( , )i x j yA
11 12 1
21 22 2
1 2
.. ..
.. ..
.. .. .. .. ..
.. .. .. .. ..
.. ..
m
m
m m mm
A A A
A A A
A
A A A
Value added decomposition – domestic and export
Input-Output Method and Leontief Inversey = A y +d => y = (I - A)-1d
Decompose total production into two parts
y = (I - A)-1d = (I - A)-1 (di + Σ j ≠ i dj) = (I - A)-1di + {(I - A)-1 Σ j ≠ i dj } = yi + y-i
VAE= value-added embedded in the production in order to satisfy final foreign demand
Value added exports of activities
Value-added Export of each industry
Decompose the industrial value-added export into the contribution by low- med-, and high-skilled labour, according to their wage shares (share of wage earned by each kind of labour as a percentage of total value-added in this industry)
Relatedness between each pair of activities can be calculated using the method mentioned aboveTHERE MAY BE SOME NOTATIONAL ISSUES BUT NOT CRITICAL TO THE PAPER?
Measure of Revealed Comparative Advantage (RCA)
Value-added export RCA of activity x in country a in year t is given by
A value of RCA greater than 1 means the country a has comparative advantage in activity x.
Borrowed from Hidago et al (2007) Relatedness between two activities x and y is measured using the
conditional probability of having comparative advantage in y, given that the country already has comparative advantage in x.
This is computed in the paper using:
Notice that this measure is not symmetric. A symmetric measure is given by
Concept of “relatedness” between activities
1 1 1 1 , , ,x ,x ,ymin Prob | ;Prob |x y a y a a aRCA RCA RCA RCA
Heatmap of relatedness index
Almost all kinds of low-skilled activities
Clusters of activities by relatedness measure
Clustering of activities
Econometric Model Factors resulting in comparative advantage in a certain activity x in country i
Dependent variable:
As the dependent variable is binary, use Probit analysis.
Explanatory variables: PRODY index
This index shows the per capita income of a representative exporter of x. Activities with higher PRODY indices are those that have high export shares in rich countries Maximum Link variable for activity x in country I is the maximum relatedness of x with
those activities that country i holds a comparative advantage in 1995.
RE = Share of employment in highly related activities with related index > 0.55
09 951 1 1
0
, , , ,
,
|i x i xi x
if RCA RCAY
Otherwise
95 951
1
,, ,
,
ni x
x ini i xi
VAEPRODY y
VAE
95 1 , , ,max ; theset of activities thati x y y x i yMaxLink y RCA
Estimated Model: Baseline and Robustness checks
Main Findings
Labour supply effect Employment in highly related activities can shift into the new
activity easily, and, therefore, has the potential to supply labour for the new activity
Agglomeration and Economy of Scope Highly related activities have backward/forward linkages with the
new activity The existence of a large size of related activities increase the return
for the new activity, therefore attract new investments Distinguish between the two effects
For the labour supply effect: labour would shift towards a new activity when it is more desirable than the current activity. For agglomeration effect, there is no such “additional requirement”
Divide the share of employment in highly related activities into two parts: those activities that have higher/lower potential for growth (i.e. PRODY) and to the new activity
Conclusions
Under the possibility of offshoring, the structural upgrading should be viewed as the skill upgrading in actual activity, rather than in industry or product.
Movement from low-skilled to high-skilled activities is difficult, horizontal policies like education and training arrangements might work better than industrial policies (e.g. stimulating on some high-tech industries)
The relatedness indices and empirical paths found in this paper have significant predictive power on actual direction of structural change
Discussion/Comments
■ It is an impressive paper with a lot of work involved.■ The paper is innovative in using valued added based measure of revealed comparative
advantage instead of the traditional measure based on export performance.■ The paper demonstrates one of many applications for the WIOD data – a novel
application to study structural change and development.■ My comments are on the conceptual framework.
■ The measure of relatedness focuses on the event RCAa,y > 1 conditional on RCAa,x > 1. In a temporal sense, the relatedness could relate to improvements on RCAa,y conditional on advantage in activity x.
■ The estimate on page 7 is computed as the ratio on the number of countries with RCAa,y >1 and RCAa,x > 1 to the number of countries that have RCAa,x > 1.
■ This estimate cannot be based on data from a single year as the number of countries covered is small.
■ The paper pools the data over the years 1995 to 2007 to derive estimated probabilities. But this could be problematic because RCA data over time are not a random sample but instead correlated over time.
Discussion/Comments
■ Can the concept of RCA be improved by looking at only activities y belonging to higher skill set compared to that of skill set x? This could be the key to study structural change.
■ As the RCA concept is directional and also the extent of improvement is of interest, it would be good to see if weighted directional graphs could be used to represent the
state of play. ■ From the heat-map and the network graph, it is clear that there is a lot of activity
in the block-diagonals, i.e. within the same labour skill set but across different industries.
■ In this case the event of interest should be the relatedness between activities across different skill sets
■ The paper does not make this distinction. Would this be useful to study structural change?
Discussion/Comments
■ The paper makes use of a probit model to identify the factors that lead to the event RCA09 > 1 given that RCA05 > 1 .
■ This model places heavy emphasis on the cut-off of 1. A better approach would be to model improvements in RCA using a Tobit-model with dependent variable taking value 0 if there is no improvement or deterioration and the actual level of improvement when such an improvement takes place.
■ Given that RCA measures are likely to be correlated over time, it is necessary to allow for possible autocorrelation in the model
■ Also evident from the relatedness graph, there could be strong contemporaneous correlation present which may need to be considered in econometric estimation.
■ How can this be done? I have no clue, but it should be possible.
Enjoyed reading the paper! Thank you!