Economics Letters 97 (2007) 230 – 234

www.elsevier.com/locate/econbase

Relative backwardness, absorptive capacity

and knowledge spillovers
Rod Falvey a , Neil Foster b , David Greenaway a,⁎
a
University of Nottingham, UK
b
University of Vienna, Austria

Received 2 December 2005; received in revised form 13 February 2007; accepted 15 March 2007
Available online 5 July 2007

Abstract

Panel data are used to investigate North–South trade-related knowledge spillovers. We find that absorptive
capacity increases the benefits of knowledge spillovers, and that spillovers have least impact in countries closest to
and farthest from the technological frontier.
© 2007 Elsevier B.V. All rights reserved.

Keywords: Economic growth; Knowledge spillovers

JEL classification: F43; O30; O40

1. Introduction

Evidence of knowledge spillovers through trade has been found among developed countries and
from developed to developing countries. Despite interest in understanding the role of domestic factors
in determining the extent of spillovers little evidence on this exists. Here we examine the dependence
of North–South spillovers on domestic factors, specifically relative backwardness and absorptive
capacity.

⁎ Corresponding author. School of Economics, University of Nottingham, Nottingham NG7 2RD, UK. Tel.: +44 1159515469;
fax: +44 1159515552.
E-mail address: [email protected] (D. Greenaway).

0165-1765/$ - see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.econlet.2007.03.015
 R. Falvey et al. / Economics Letters 97 (2007) 230–234 231

2. Background

Coe et al. (1997) (CHH) identify several channels whereby foreign knowledge can be transferred
including: imports of intermediate and capital goods; cross-border learning of production methods,
product design and organization; imitation of new products; development of technologies and imitation of
foreign technology. These arguments underlie tests of links between knowledge spillovers through trade
and output or productivity growth. In a seminal paper Coe and Helpman (1995) (CH) test for trade-related
knowledge spillovers (TRKS) among 22 OECD countries for 1971–1990. A stock of knowledge is
constructed for each country with access measured by weighting stocks with trade flows. They conclude
that both foreign and domestic stocks are important for productivity growth, with more open economies
gaining most. CHH examine North–South spillovers from these countries to 77 developing countries and
find North–South TRKS are substantial. These studies have generated a literature broadly confirming the
growth-enhancing effects of TRKS.
While the role of trade has been examined extensively, domestic factors have not. Two factors stand
out: relative backwardness and absorptive capacity. It is often argued that more backward countries have a
larger stock of foreign technology to borrow from and therefore a larger potential for foreign knowledge
spillovers (Gerschenkron, 1962). However, Matthews (1969) argues that proximity to the technological
frontier may make it easier to borrow foreign technology. The size of the technology gap alone may not
determine the magnitude of spillovers, with the ability to adapt such knowledge – absorptive capacity –
being important (Abramowitz, 1986). Measures of backwardness are typically included as “catch-up”
terms in growth regressions. When included with TRKS, such terms capture knowledge spillovers
through other channels and results generally confirm advantages of backwardness. Interacting educational
attainment with a spillover variable is one test of the absorptive capacity hypothesis, and evidence
indicates that this enhances advantages of backwardness (Benhabib and Spiegel, 1994; Engelbrecht,
1997), but not TRKS (CHH, Engelbrecht, 1997).

3. Empirical specification and results

We concentrate on spillovers from five R&D intensive OECD ‘donor’ economies to 57 developing
‘recipient’ economies1 , using panel data on five non-overlapping 5-year periods between 1975 and 1999.
Our empirical specification relates the growth rate of output per worker (ΔGDPLrt) to the growth of
TRKS (ΔTRKS) and other growth related variables. The initial equation is:
DGDPLrt ¼ b1 GAPrt þ b2 INVrt þ b3 DLAB þ b4 SYRrt þ b5 TRADE þ b6 GAPrt  SYRrt
þ d1 DTRKSrt þ lr þ tt þ ert
where INVrt is the ratio of investment to GDP (for recipient r in time period t), ΔLAB is labor force
growth, SYR our measure of schooling, TRADE the average ratio of imports plus exports to GDP
1
Donors are France, Germany, Japan, UK and USA. Developing countries are: Algeria, Argentina, Bangladesh, Bolivia,
Brazil, Cameroon, Central African Republic, Chile, Columbia, Costa Rica, Dominican Republic, Ecuador, El Salvador, Ghana,
Guatemala, Guyana, Haiti, Honduras, Hong Kong, India, Indonesia, Iran, Israel, Jamaica, Kenya, South Korea, Malawi,
Malaysia, Mauritius, Mexico, Mozambique, Nepal, Nicaragua, Niger, Pakistan, Panama, Paraguay, Peru, Philippines, Senegal,
Sierra Leone, South Africa, Sri Lanka, Sudan, Syria, Tanzania, Thailand, Togo, Trinidad and Tobago, Tunisia, Turkey, Uganda,
Uruguay, Venezuela, Zaire, Zambia and Zimbabwe. The maximum number of observations is 268 due to missing data for some
countries.
232 R. Falvey et al. / Economics Letters 97 (2007) 230–234

Table 1
Initial results and interactions
ΔGDPL 1 2 3 4 5 6
GAP 0.27 (5.08)⁎⁎⁎ 0.46 (5.40)⁎⁎⁎ 0.33 (3.76)⁎⁎⁎ 0.34 (3.61)⁎⁎⁎ 0.32 (3.66)⁎⁎⁎ 0.31 (3.34)⁎⁎⁎
INV 0.22 (4.82)⁎⁎⁎ 0.24 (4.96)⁎⁎⁎ 0.19 (4.12)⁎⁎⁎ 0.19 (4.11)⁎⁎⁎ 0.20 (4.26)⁎⁎⁎ 0.20 (4.25)⁎⁎⁎
ΔLAB − 0.24 (−0.66) − 0.15 (−0.45) −0.27 (− 0.76) −0.26 (− 0.73) −0.26 (− 0.74) −0.26 (− 0.76)
SYR − 0.004 (−0.56) 0.04 (2.34)⁎⁎ 0.02 (1.41) 0.03 (1.44) 0.02 (1.44) 0.02 (1.28)
TRADE 0.02 (1.92)⁎ 0.02 (1.79)⁎ 0.02 (1.33) 0.0 (1.31)2 0.01 (1.11) 0.01 (1.09)
GAPrt × SYRrt − 0.06 (−2.86)⁎⁎⁎ −0.03 (− 1.53) −0.03 (− 1.56) −0.03 (− 1.65) −0.03 (− 1.51)
ΔTRKS 0.07 (4.53)⁎⁎⁎ 0.12 (1.16) 0.05 (1.85)⁎ −0.003 (− 0.03)
ΔTRKS × GAP −0.05 (− 0.45) 0.05 (0.45)
ΔTRKS × SYR 0.03 (1.27) 0.03 (1.29)
F-Stat 7.21⁎⁎⁎ 8.54⁎⁎⁎ 10.59⁎⁎⁎ 10.67⁎⁎⁎ 11.02⁎⁎⁎ 10.84⁎⁎⁎
R2 0.61 0.62 0.66 0.66 0.66 0.66
Notes: All equations include a full set of unreported country and time dummies. t-Statistics in brackets. All models estimated
using white heteroscedasticity-consistent standard errors. ⁎, ⁎⁎, ⁎⁎⁎Indicate significance at the 10, 5 and 1% levels respectively.

(capturing other benefits of openness), μi and νt are country and time fixed effects respectively and εrt
the error term. Our measure of relative backwardness is the proportional difference of initial GDP per
worker in the recipient from that in the US: 2

INITGDPLrt  INITGDPLUSt
GAPrt ¼
INITGDPLUSt

When included linearly, backwardness accounts for other forms of catch-up including disembodied
knowledge spillovers. Although the multidimensional nature of absorptive capacity is difficult to capture
in a single variable, we follow Abramowitz's suggestion that education is a useful proxy, employing the
average years of secondary schooling in the population over 25 (SYR). The interaction between relative
backwardness and education (GAPrt × SYRrt) has been used in the past to account for the role of education
in the absorption of disembodied spillovers (Benhabib and Spiegel, 1994). We define

Mrt X 5
Sdt
TRKSrt u hdrt ;
GDPrt d¼1 GDPdt

where GDPrt (GDPdt) is the level of GDP in the recipient (donor), Mrt the recipient's imports of
machinery and transport equipment from the 5 donors, θdrt is the share of donor d in Mrt, and Sdt the
donor's stock of knowledge.3
Outcomes for the core variables in our base specification (column 1 of Table 1) are in line with previous
results. GAP is positive and significant, so the more backward a country the higher its growth. Our
measure of education is insignificant, which is common in the literature. When the interaction between
schooling and backwardness is added (column 2), its coefficient is negative and significant, contradicting
the conclusions of Benhabib and Spiegel (1994) that more schooling can enhance the advantages of

2
Using the difference from the average of our five donors gives similar results on our main variables of interest.
3
Knowledge stocks were constructed using the perpetual inventory method with a 5% depreciation rate.
 R. Falvey et al. / Economics Letters 97 (2007) 230–234 233

Table 2
Threshold results
ΔGDPL Single threshold Double threshold
GAP SYR GAP/GAP
GAP 0.32 (3.70)⁎⁎⁎ 0.34 (3.83)⁎⁎⁎ 0.32 (3.68)⁎⁎⁎
INV 0.19 (4.17)⁎⁎⁎ 0.19 (4.28)⁎⁎⁎ 0.19 (4.11)⁎⁎⁎
ΔLAB −0.27 (−0.77) − 0.32 (−0.95) −0.28 (− 0.84)
SYR 0.03 (1.54) 0.03 (1.63) 0.02 (1.34)
TRADE 0.02 (1.23) 0.02 (1.28) 0.02 (1.29)
GAPrt × SYRrt −0.03 (−1.68)⁎ − 0.04 (−1.76)⁎ −0.03 (− 1.45)
ΔTRKS
δ1 (THrt ≥ λ1) 0.10 (4.69)⁎⁎⁎ 0.04 (1.79)⁎ 0.07 (2.60)⁎⁎
δ2 (THrt N λ1) 0.05 (2.78)⁎⁎⁎ 0.09 (5.29)⁎⁎⁎
δ21 (λ1 ≤ THrt ≤ λ2) 0.14 (4.87)⁎⁎⁎
δ22 (THrt N λ2) 0.05 (2.82)⁎⁎⁎
Threshold 1 0.961 (60th) 0.523 (24th) 0.894 (37th)
Threshold 2 0.961 (60th)
Bootstrapped p-value 0.048⁎⁎ 0.047⁎⁎ 0.078⁎
F-Stat 11.67⁎⁎⁎ 10.85⁎⁎⁎ 12.03⁎⁎⁎
R2 0.66 0.66 0.67
Notes: As in Table 1. Thresholds 1 and 2 report the estimated first and second thresholds along with the percentile of the
distribution of the estimated threshold. Bootstrapped p-values report the results of the bootstrap test of the significance of the
estimated threshold.

backwardness. However, the addition of TRKS (column 3) renders both the general trade measure and the
interaction between schooling and backwardness insignificant, suggesting that these variables were
capturing some of the benefits due to embodied knowledge spillovers. The coefficient on TRKS is similar
to those in CHH.4 The final three columns include interactions between TRKS and GAP and SYR, but
find little evidence of a role for backwardness or schooling in enhancing the benefits of TRKS.5
An alternative approach to modeling non-linearities involving TRKS, is Hansen's (1999) threshold
regression method for panel data, where observations fall into regimes dependent upon an unknown value
of an observed variable. This allows the coefficient on TRKS to depend on GAP or SYR. For a single
threshold the estimating equation is:
DGDPLrt ¼ b1 GAPrt þ b2 INVrt þ b3 DLAB þ b4 SYRrt þ b5 TRADE þ b6 GAPrt  SYRrt
þ d1 DTRKSrt IðTHrt Vk1 Þ þ d2 DTRKSrt IðTHrt Nk1 Þ þ lr þ tt þ ert
where THrt is the threshold variable, I(·) is an indicator variable and λ1 is the estimated threshold.
Here observations are allocated to regimes depending on whether the threshold variable is smaller or
larger than λ. The impact of TRKS on growth will be δ1 (δ2) for countries in the low (high) regime.6
The threshold value λ and the slope parameters are jointly estimated and then the null (δ1 = δ2) is tested.7

4
Our results suggest that a 1% increase in foreign R&D increases growth by 0.07% compared with CHH finding 0.06%.
5
Tests of the combined significance of variables included in linear and interactive forms show that, while GAP and TRKS are
significant, schooling tends to be insignificant.
6
To ensure a reasonable number of observations we require at least 20% must be in each regime.
7
Using bootstrap procedures described in Hansen (1999) with 1000 replications.
234 R. Falvey et al. / Economics Letters 97 (2007) 230–234

Results are reported in Table 2, where core variables are largely as before. Estimating a single threshold
on both GAP (column 1) and SYR (column 2) results in significant thresholds at the 60th and 24th
percentiles, respectively. The coefficients on TRKS are larger in the low regime for GAP, and in the high
regime for SYR. The latter implies that countries with higher absorptive capacity gain more from TRKS.
The results on GAP imply that countries closer to the technological frontier gain more from TRKS, a
result supportive of the argument of Matthews (1969) reported above.
Further thresholds can be considered by fixing the first threshold at its estimated value and searching for a
second threshold, λ2.8 The results suggest only a single threshold for SYR, but we do find evidence of a
significant second threshold for GAP.9 The coefficients indicate that backwardness increases the benefits of
TRKS, with the largest benefits to countries in the middle regime (between the 37th and 60th percentiles).

4. Conclusions

Our results confirm that TRKS are an important source of growth for developing countries. Threshold
analysis suggests that countries benefit from TRKS at all education levels, but those with the highest
levels of education benefit more. It also confirms the advantages of backwardness for TRKS, but suggests
a role for absorptive capacity. The largest spillovers accrue to countries that are not too backward but still
sufficiently far from the frontier that significant foreign knowledge remains unexploited.

Acknowledgements

The authors acknowledge helpful comments on an earlier draft from an anonymous referee and Eric
Maskin. Falvey and Greenaway gratefully acknowledge financial support from The Leverhulme Trust
under Programme Grant F 114/BF.

References

Abramowitz, M., 1986. Catching-up, forging ahead, and falling behind. Journal of Economic History 385–406.
Bai, J., 1997. Estimating multiple breaks one at a time. Econometric Theory 13, 315–352.
Benhabib, J., Spiegel, M.M., 1994. The role of human capital in economic development: evidence from aggregate cross-country
data. Journal of Monetary Economics 34, 143–173.
Coe, D.T., Helpman, E., 1995. International R&D spillovers. European Economic Review 37, 557–565.
Coe, D.T., Helpman, E., Hoffmaister, A.W., 1997. North–South R&D spillovers. Economic Journal 107, 134–150.
Engelbrecht, H.-J., 1997. International R&D spillovers, human capital and productivity in OECD economies: an empirical
investigation. European Economic Review 41, 1479–1488.
Gerschenkron, A., 1962. Economic Backwardness in Historical Perspective. Belknap Press, Cambridge MA.
Hansen, B.E., 1999. Threshold effects in non-dynamic panels: estimation, testing and inference. Journal of Econometrics 93,
345–368.
Matthews, R.C.O., 1969. Why growth rates differ. Economic Journal 79, 261–268.

8
Applying the refinement estimator of Bai (1997) (i.e., re-estimating the first threshold fixing the second threshold at its
estimated value), did not change the estimated first thresholds. The significance of the second threshold was again tested using
the bootstrap procedures discussed in Hansen (1999).
9
No significant third threshold for GAP nor any significant cross-thresholds (i.e., a second threshold based on GAP (SYR)
given the first threshold based on SYR (GAP)) were found.