The Penn effect and transition: the new EU member states in international perspective.
Frensch, Richard ; Schmillen, Achim
INTRODUCTION AND MOTIVATION
Aggregate price levels expressed in a common currency at going
nominal exchange rates are generally higher in richer than in poorer
economies, an observation dubbed the 'Penn effect' by
Samuelson (1994). Recent panel data studies (eg, Dobrinsky, 3003; de
Broeck and Slok, 2006) have found comparatively high point estimates for
corresponding price-productivity elasticities for transition economies
in Central and Eastern Europe (CEEC), which seems to contrast with
findings from cross-section regression analyses, where the inclusion of
poorer countries tends to generate lower elasticities (Maeso-Fernandez
et al., 2005).
Whether or not a special status for transition economies exists in
the Penn effect appears important for two reasons. First, it might lead
to higher observed inflation in these countries, which the ECB, among
others, could be concerned about. Second, it might lead to losses in
competitiveness, which policy makers in these countries should be
concerned about.
The idea of this paper is to put the price-productivity behaviour
of transition economies into international perspective. For this
purpose, we first review the literature on the Penn effect. This
literature seems to show a special status for transition economies. In
the subsequent section, we demonstrate that within the time-series
dimension, estimated price-productivity elasticities for transition
economies are indeed different from those of non-transition OECD
countries. In the following, however, we argue that (i) the Penn effect
is by its nature a cross-section rather than a time-series phenomenon,
and (ii) that estimates of price-productivity elasticities without the
inclusion of other explanatory factors might suffer from omitted
variable bias and omitted variable inconsistency. As a solution, we
propose an extended approach in order to take account of reform effort
as the driving force behind deregulation, reallocation, and
restructuring during transition. The results of estimating the extended
approach with panel data and fixed period-effects suggest that the
price-productivity elasticity for transition economies is not different
from that of OECD economies.
THE PENN EFFECT FOR TRANSITION ECONOMIES
Purchasing power parity (PPP) is linked to the tradability of goods
and services. If all goods are tradable at no cost and enter each
country's aggregate price level basket with the same weight,
arbitrage ensures that aggregate price levels, [P.sub.1] and [P.sub.2],
are identical for each pair of countries when expressed in a common
currency at the going nominal exchange rate. More generally, the
deviation of the nominal exchange rate [e.sub.12] from purchasing power
is just the real exchange rate between countries 2 and 1, [RER.sub.21].
[RER.sub.21] = [e.sub.12]/[P.sub.1]/[P.sub.2] =
[P.sub.2]/[P.sub.1]/[e.sub.21] (1)
equivalently defined as the deviation of the ratio of two
countries' aggregate price levels from their nominal exchange rate.
Absolute PPP, of course, is equivalent to [RER.sub.21] = 1.
In fact, what we observe are systematic deviations from PPP:
aggregate price levels expressed in a common currency at going nominal
exchange rates are generally higher in richer than in poorer economies,
an observation dubbed the Penn effect by Samuelson (1994).
By far the most prominent explanation for the Penn effect is the
Balassa-Samuelson (BS) hypothesis (Balassa, 1964; Samuelson, 1964).
Balassa and Samuelson rationalise the effect in a chain of arguments
building on (a) purchasing power for tradables, (b) relative prices
reflecting relative labour productivities, (c) homogenous national
labour markets across sectors of production, and (d) overwhelming
differences in labour productivity across countries to be found in
tradable rather than in non-tradable production. (1) Leaving (d) aside
defines the productivity gap version of the BS hypothesis: the real
exchange rate between each pair of countries 2 and 1 is the higher, the
higher country 2's ratio between its tradables and non-tradables
sector productivities compared with country 1:
[RER.sub.21] = [P.sub.2]/[P.sub.1]/[e.sub.21] =
[[[([A.sup.T.sub.2][A.sup.N.sub.1]).sup.1 - [theta]]]/
[[([A.sup.T.sub.1][A.sup.N.sub.2]).sup.1 - [theta]]]] (2)
where [A.sup.T] and [A.sup.N] are labour productivities in tradable
and non-tradable sectors, and equal preferences across countries are
described by constant and equal consumption expenditure shares for
tradables and non-tradables, [theta] and 1 - [theta], respectively. (2)
Adding observation (d), that is, that cross-country productivity
differences are concentrated in the tradable goods sector, immediately
implies the Penn effect: for each pair of countries, their real exchange
rate is a positive function of their ratio of overall productivities,
with the consumption expenditure share for non-tradables corresponding
to the elasticity of the real exchange rate with respect to relative
productivity (the price-productivity elasticity):
[RER.sub.21] = [P.sub.2]/[P.sub.1]/[e.sub.21] =
[([y.sub.2]/[y.sub.1]).sup.1 - [theta]] (3)
Empirical work on the Penn effect like Bergstrand (1991), Lothian
and Taylor (2008) or Chong et al. (2012) typically studies relationships
between countries' multilateral real exchange rate measures and
productivities. The most popular measures of countries'
multilateral real exchange rates are (i) effective real exchange rate
indices, that is, weighted sums of each country's bilateral nominal
exchange rates deflated by consumer price indices with weights
corresponding to the relative importance of partner countries in trade;
or (ii) comparative prices (or exchange rate gaps in much of the
literature), as provided in the Penn World Tables (PWT), defined as the
deviation of a country's nominal exchange rate against the
international dollar from purchasing power.
Each country's comparative price level is, by construction, a
weighted real exchange rate against the international dollar, where the
weighting scheme is based on the relative prices that underlie the
derivation of the international dollar, thus providing a measure of
Equation 1 that is conceptually close to, and highly correlated with, a
trade-weighted real effective exchange rate index. However, comparative
price levels have the enormous advantages of being more widely available
and of being internationally comparable in level terms, which is why we
use them in the rest of this paper. (3)
Figure 1 displays the benchmark price-productivity relationship for
a number of OECD and transition economies between 1992 and 2004, with
average productivity proxied by PPP-adjusted income per capita. The
literature on transition countries has, with the notable exceptions of
de Broeck and Slok (2006), Frensch (2006), and Garcia-Solanes et al.
(2008), so far been mostly confined to identifying Penn or BS effects
within this country group's data, without putting them into an
international perspective. Early results in this vein had been used as a
basis for arguing that real appreciation in the region is to a large
extent due to the BS effect (Halpern and Wyplosz, 2001). However, Egert
and Halpern (2006) in their meta-regression analysis of studies of CEEC
real exchange rates fail to find a significant influence of a simple
BS-driven behaviour on real exchange rate developments in the region.
Recent work has supported this view on the ground that, even for
tradables, PPP need not necessarily hold, for example, due to a quality
adjustment bias (cf. Cincibuch and Podpiera, 2006). (4) Egert et al.
(2006) stress three stylised facts of real exchange rate behaviour in
transition:
(1) Until around the mid-1990s transition countries'
currencies were substantially undervalued in terms of PPP.
(2) Different from the BS pattern of explanation of the Penn
effect, all types of goods, not only non-tradable services, were or
still are undervalued in terms of PPP.
(3) Different in extent across countries, the region has witnessed
strong appreciation from the outset of transition.
[FIGURE 1 OMITTED]
Accordingly, the possibility of a special relationship between
productivity and aggregate price levels for transition economies,
evident from Figure 2, arises because aggregate price levels of a former
centrally planned economy (CPE) may ceteris paribus be biased downwards:
price liberalisation may still be incomplete, that is, the output of a
former CPE is not yet fully priced on the market, subsidisation drives a
wedge between prices and costs especially for services, that is,
non-tradables. Moreover, output quality is systematically lower in a
former CPE than in a market economy (Frensch, 2004; Garcia-Solanes et
al., 2008). On the other hand, a number of transition countries,
especially in the CIS, are oil and gas exporters where related Dutch
disease phenomena might drive up comparative prices. In terms of a
theoretical foundation, Clague (1985) proposes that, within a
specific-factors model, increases in the endowment of specific factors,
one of which is natural resources, lead to higher comparative price
levels, as do productivity increases.
[FIGURE 2 OMITTED]
ESTIMATION AND RESULTS
The time-series dimension
One drawback of using panel data lies in the potential
non-stationarity of price and productivity data. This is of specific
concern with panels too short for proper panel unit root testing. De
Broeck and Slok (2006, pp. 377-378) employ Pooled Mean Group (PMG)
estimations for the long-run time series dimension of the relationship
between productivity and real effective exchange rates stating 'in
case the variables are 1(1), estimation is conducted under the untested
assumption that there exists a long-run relationship such that the error
term in the estimated long-run equation is stationary'. However,
their procedure is not completely without problems: PMG estimations
focussing on the time-series dimension are done with a very short panel
(1991-1998) and are derived only for CEEC and CIS countries and not
estimated for OECD countries or any other control group. (5)
On the choice between fixed effects and alternative estimators for
potentially non-stationary data, Fidrmuc (2009) in the gravity context
uses cross-sectionally augmented panel unit root testing methods and
confirms that trade and income variables used in gravity regressions are
integrated of order one. However, Fidrmuc (2009, p. 436) finds that,
although fixed effects estimators may be biased, they are not only
asymptotically normal and consistent with large panels but also perform
'relatively well in comparison to panel cointegration
techniques' in finite samples, concluding the potential bias of
fixed-effects gravity estimators to be rather small.
Accordingly, we start by analysing the time-series dimension of the
Penn effect in a panel OLS regression with country fixed effects to
control for plausibly important time-invariant country-specific
unobserved heterogeneity with the implication that no time-invariant
influences can be estimated. Data on PPP-adjusted income per capita, y,
to proxy average productivity and p are taken from the PWT, version 6.2
(see notes to Figure 1). The data cover 41 countries (ie, 12 CEEC, 9
CIS, and 18 non-transition OECD, see Table B1) over 1992-2004, resulting
in a panel size of 484 observations. (6)
The results reported in column 1 of Table 1 confirm a significant
benchmark Penn effect with special status for transition economies; in
particular, we note a negative price-productivity relationship for CIS
economies. For CEEC the coefficient is positive but, in contrast to what
is found by recent panel data studies such as Dobrinsky (2003) and de
Broeck and Slok (2006), not statistically significant.
While the special status of transition economies is routinely
explained by reform efforts, or lack thereof, in these countries
influencing productivity growth, the relevant literature does not
directly include reform variables in its estimations. As a consequence,
it is in fact unable to identify reform effects. What is more,
structural reforms are likely to jointly influence p and y. So their
omission entails an omitted variable problem, with y being endogenous
and its estimated coefficient potentially biased and asymptotically
inconsistent. More specifically, reforms in Central and Eastern European
transition economies can be expected to have pushed up both productivity
and prices in these countries (cf. Dufrenot and Egert, 2005). Thus, the
particularly high price-productivity elasticity in the transition
context reported by much of the literature could be partly or even
entirely due to the omission of reform variables. The reverse might be
the case for the CIS economies.
In Appendix A, we exemplify a simple extension to the static
BS-based approach to the Penn effect, focussing on real factors and
reforms like Coricelli and Jazbec (2004) and Garcia-Solanes et al.
(2008). According to this extended approach, real exchange rate
developments react to productivity developments, reform-driven quality
improvements and sectoral reallocation and the competition effect of
trade liberalisation. (7) While trade liberalisation and competition are
per se reform variables, all other variables are also influenced by
various reform efforts, and potentially dominated by them, in particular
in transition economies. Rather than attempting a structural estimation,
we take this extended BS approach as motivation to estimate
price-productivity elasticities by controlling for reform effort.
A priori, we would expect price liberalisation, that is, lessening
of administrative price controls, to imply higher price levels, given
prevailing shortages at the outset of transition. (8) In similar vein,
trade and foreign exchange system liberalisation would have the same
effect, while competition policy should ceteris paribus have a price
decreasing effect. Small-scale privatisation can be expected to be
linked to positive price effects because private rather than state
provision of private goods is linked to cost coverage. This mechanism
should also be present for large-scale privatisation. However, as
Hanousek and Kocenda (2010) show, large-scale privatisation often goes
hand in hand with disinvestments or the outright break-up of
conglomerates, which might lead to lower prices. Therefore, a priori,
the overall effect of large-scale privatisation on prices is uncertain.
The EBRD Transition Indicators measure reform progress along
several dimensions, in terms of price liberalisation, trade and foreign
exchange system liberalisation, competition policy, large-scale
privatisation, and small-scale privatisation on a scale with one-third
steps between 1 and 4.33. We assume these indices to equal 4.33 for OECD
economies, in line with their construction (cf. Table B2). While the
EBRD transition indicators are often used as cardinal measures, they are
probably ordered qualitative rather than cardinal and should not be used
directly in linear regression analysis. For this reason, we construct
dummy variables from these indicators in the general form of
[ReformMeasure_Level.sub.j,t], indicating whether or not country j has
within a certain policy field made the step towards a certain level on
the EBRD scale at some point in time. With reform progress measured in
steps of one-third of a point, quite a number of dummy variables are
conceivable. Specifically, we construct dummy thresholds at median value
for transition countries to assess reform impact on comparative prices
(for more on this, see the next section on sensitivity).
The results reported in column 2 of Table 1 confirm the existence
of a benchmark Penn effect in the time-series dimension. Much of the
column 1 special status for transition economies is now picked up by the
transition indicators broadly in line with a priori expectations.
Competition policy, however, exhibits an insignificant coefficient.
The regressions reported in columns 1 and 2 of Table 1 are
problematic because the focus on the within-variation of the
price-productivity relationship might aggravate measurement errors in
the PPPs defining p, much of which is essentially unobservable. Between
ICP rounds, changes in data and methods are regularly introduced. (9)
Furthermore, available reform variables show rich between- but little
within-country variation, which is especially true for structural
reforms such as progress with competition policy. The inclusion of
country fixed effects implies that no time-invariant parameters, such as
potentially important natural resource endowments, can be included in
the regression. Controlling for time-invariant country-specific
unobserved heterogeneity makes it difficult to motivate y as a good
proxy for productivity in a world of synchronised business cycles.
The cross-section dimension
As stated above and forcefully argued by Samuelson (1994) and
Bergin et al. (2006), the Penn effect is fundamentally a cross-section
phenomenon: aggregate price levels expressed in a common currency at
going nominal exchange rates are generally higher in richer than in
poorer economies.
In line with this, two strands of empirical literature suggest that
a closer look at the cross-section dimension of this relationship might
indeed be revealing. First, Maeso-Fernandez et al. (2005, p. 139) report
that price-productivity elasticity estimates from cross-section
regressions vary greatly with sample composition. '(T)he inclusion
of poor countries--particularly, African countries tends to generate
lower elasticities'. The evidence in Frensch (2006) also suggests
different strengths of the p-y relationship in sub-samples of countries,
with an especially pronounced relationship for OECD countries. (10)
Second, Bergin et al. (2006, p. 4) conclude that in a sequence of
PWT cross-sections every 5 years between 1950 and 1995, the relationship
has gradually strengthened, 'with the slope estimate roughly
quadrupling in size over half a century'. Why the Penn effect has
strengthened over time remains a question of active research. According
to one straightforward explanation rooted in the underlying BS effect,
the consumption expenditure share for nontradables than might have
increased over time. However, in fact, in 1950 traded shares of output
were lower than both in 1913 and in 2000 (Taylor and Taylor, 2004).
Rather, recent approaches to endogenise BS effects (see especially
Bergin et al., 2006) start out with the hypothesis that declining trade
costs increase tradability, such as in models of heterogeneous firms and
trade (Melitz, 2003). (11)
Because of the relatively small number of cross-sectional
observations, however, we are unable to thoroughly explore the between
variation of the price-productivity relationship. We rather compromise
by running a panel OLS regression with period-fixed effects; this
controls for plausibly important time-specific country-invariant
unobserved heterogeneity with the implication that synchronised business
cycles are captured to better proxy productivity with PPP-adjusted per
capita income, y.
In addition to the data used in the previous section, the IMF Guide
on Resource Revenue Transparency (2007) is used as a source for dummies
for hydrocarbon-rich countries.
The results reported in column 3 of Table 1 confirm the existence
of a cross-country benchmark Penn effect, quantitatively close to the
one found for the time-series dimension. Most notably, there is no
special status for transition economies; rather, dummies for
hydrocarbon-rich countries based on IMF (2007) and some transition
indicators play an important role. In particular, price and trade and
foreign exchange system liberalisation imply higher price levels; the
same holds for privatisation, although not significantly so for
large-scale privatisation. Competition policy is again not associated
with comparative price levels in a statistically significant way.
SENSITIVITY
Choice of sample
The major qualitative results of the previous section are the
existence of a Penn effect implying a price-productivity elasticity of
about 0.5 and the finding that, within our cross-section specification,
there is no special status for transition economies. Rather,
identifiable time-varying country-specific variables such as energy
dependence and the extent of reforms ceteris paribus have a significant
influence on aggregate price levels. These results are robust to
excluding Armenia, Azerbaijan, Kyrgyzstan, for which we have very few
observations, from the sample, or for extending the sample period to
1989-2004 (cf. columns 4 and 5 of Table 2).
Variable definition
Our major results are also quite robust to variations in variable
definitions: we first experiment by changing the definition of oil and
gas exporters to net energy exporters as listed by IEA (2008). Second,
we vary the exact threshold for the definition of the transition
indicator dummies: rather than construct dummies at median value for
transition countries, we construct dummies by discriminating between
first tercile of transition countries versus the other two, or between
the first two terciles of transition countries versus the third. Again,
we never find a significant special status for transition economies with
the exception of the CEEC economies in our last specification (see Table
2, column 8). (12)
CONCLUSIONS
We find a robust and stable Penn effect over all our
specifications, with an implied price-productivity elasticity of about
0.5. Within the pure time-series dimension, our results confirm earlier
findings reporting the existence of a special status for transition as
compared with OECD economies. However, we argue that (i) the Penn effect
is fundamentally a cross-section phenomenon and (ii) the omission of
real factors connected to reform effort might lead to omitted variable
bias and omitted variable inconsistency. In our preferred specification,
which treats the Penn effect as a cross-section phenomenon and in which
resource dependence and the extend of reforms are included as additional
control variables in order to take account of possible endogeneity of
the productivity variable, there is no special status for transition
economies. These results are very robust with respect to choice of
sample and variable definition.
APPENDIX
An extended static BS framework for motivating Penn effects in
transition
In the simple set-up of the section 'The Penn effect for
Transition Economies', the only alternative to a deepening
productivity gap to imply a more pronounced BS-type relationship is by a
rise in the share of non-traded goods in GDP, which seems heavily at
odds with empirical developments. The argument in Frensch (2000 and
2006), on which we build here, however, allows us to separate
tradability from reallocation in terms of changes in income shares spent
on services and industrial goods. For further analysis, we return to the
arbitrage view of the BS set-up, extending the framework to incorporate
the effects of transition, defined as institutional reform-driven
resource reallocation, corporate restructuring, and liberalisation
(Blanchard, 1997). Then,
ln [RER.sub.21] = ln [P.sub.2] - ln [P.sub.1] - ln [e.sub.12] (A.
1)
following the notation in section 'The Penn effect for
Transition Economies' omitting time. Rather than differentiating
only between tradables and nontradables, we assume two sectors, industry
(I) and services (S), with products entering price levels with
potentially different weights such that,
ln [P.sub.j] = [[phi].sub.j] ln [P.sup.I.sub.j] + (1 -
[[phi].sub.j]) ln [P.sup.S.sub.j] (A.2)
We make a few simplifying assumptions to modify the set-up of the
section 'The Penn Effect for Transition Economies':
(ASS. 1) While all services are non-tradable, only some part of
industrial goods, [[tau].sub.j], is tradable due to the existence of
barriers to trade, that is,
ln [P.sup.l.sub.j] = [[tau].sub.j] ln [P.sup.I,T.sub.j] + (1 -
[[tau].sub.j]) ln [P.sup.I,NT.sub.j] (A.3)
(ASS. 2) Prices are proportional to unit labour costs,
ln [P.sup.h.sub.j] = [[lambda].sup.h] + ln [w.sub.j] - ln
[A.sup.h.sub.j] (A.4)
where h = S; I, T; I, NT, w is the wage rate and A is labour
productivity, which is the same in all of industry.
(ASS. 3) Exposure to international trade increases the intensity of
competition, that is,
[[lambda].sup.S] = [[lambda].sup.I,NT] = [[lambda].sup.NT] >
[[lambda].sup.T] = [[lambda].sup.I,T] (A.5)
(ASS. 4) PPP, as usual, does not hold for non-tradables; for
tradables, PPP is restricted by quality differentials according to
ln [P.sup.I,T.sub.2] = ln [P.sup.I,T.sub.1] + ln [e.sub.12] +
[[kappa].sup.I,T.sub.21] (A.6)
where country 2 product quality of tradables,
[[kappa].sup.I,T.sub.21], is defined relative to country 1.
From (A.1) and (A.6),
ln [RER.sub.21] = (ln [P.sub.2] - ln [P.sup.I,T.sub.2]) - (ln
[P.sub.1] - ln [P.sup.I,T.sub.1]) + [[kappa].sup.I,T.sub.21] (A.7)
where (A.2) implies that
ln [P.sub.j] - ln [P.sup.I.sub.j] = (1 - [[phi].sub.j])(ln
[P.sup.S.sub.j] - ln [P.sup.I.sub.j]) (A.8)
and from (A.3)
ln [P.sup.I.sub.j] - ln [P.sup.I,T.sub.j] = (1 - [[tau].sub.j])(ln
[P.sup.I,NT.sub.j] - ln [P.sup.I,T.sub.j) (A.9)
From (A.8) and (A.9),
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (A.10)
Substituting from (A.9),
ln [P.sub.j] - ln [P.sup.I,T.sub.j] = (1 - [[phi].sub.j]) ln
[P.sup.S.sub.j] - ln [P.sup.I,T.sub.j] + [[phi].sub.j] ln
[P.sup.I.sub.j]
and from (A.3),
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (A.11)
Substituting for prices according to (A4) and collecting terms
yields
ln [P.sub.j] - ln [P.sup.I,T.sub.j] = (1 - [[phi].sub.j])(ln
[A.sup.I.sub.j] - ln [A.sup.S.sub.j]) + (1 -
[[tau].sub.j][[phi].sub.j])([[lambda].sup.NT] - [[lambda].sup.T]) (A.12)
Then, Equation A.7 implies,
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (A.13)
After total differentiation and again collecting terms, we
decompose the rate of change of the real exchange rate of country 2
relative to country 1 into four separate effects (where a [DELTA] of a
logarithmic value indicates a growth rate),
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (A.14)
Separating tradability from income shares spent on services and
industrial goods allows us to show that, in addition to the productivity
gap effect, reallocation from industry towards services in country 2,
relative to country 1 ([DELTA] [[phi].sub.2] < 0), also implies a
real exchange rate appreciation assuming that productivity in industry
is higher than in services. Also, quality improvements drive up the real
exchange rate. A unilateral reduction in country 2 versus country 1
foreign barriers to trade in industrial products ([DELTA] [[tau].sub.2]
> 0 and [DELTA] [[tau].sub.1] = 0) implies a real depreciation for
country 2. Symmetric reduction in barriers to trade ([DELTA]
[tau].sub.1] = [DELTA] [tau].sub.2] 0) implies a depreciation for
country 2 as long as the share of this country's services sector in
total production is smaller than in country 1. While this depreciation
effect is rooted in the pro-competition effect of trade liberalisation,
trade liberalisation, along with all other reform measures described in
section 'Estimation and results', influences and even
dominates restructuring efforts and sectoral reallocation, specifically
pronounced during transition.
APPENDIX B Data
Table B1: Countries covered
1 Albania*
2 Armenia**
3 Austria
4 Azerbaijan**
5 Belgium
6 Bulgaria*
7 Belarus**
g Canada
9 Switzerland
10 Czech Republic*
11 Denmark
12 Spain
13 Estonia*
14 Finland
15 France
16 United Kingdom
17 Georgia**
18 Germany
19 Greece
20 Croatia*
21 Hungary*
22 Ireland
23 Iceland
24 Italy
25 Kazakhstan**
26 Kyrgyzstan**
27 Lithuania*
28 Latvia*
29 Netherlands
30 Norway
31 Poland*
32 Portugal
33 Romania*
34 Russia**
35 Slovakia*
36 Slovenia*
37 Sweden
38 Turkmenistan**
39 Turkey
40 Ukraine**
41 United States**
Notes: CEEC countries underlined, CIS countries
in italics. Other countries are OECD as of 1992.
Note: CEEC countries indicated with *.
Note: CIS countries indicated with **.
Table B2: Variables used in regressions (1)-(8) in Tables 1 and 2
Variable Definition Source Notes
p Comparative Penn World p is the PPP over GDP
prices, measured Tables divided by the exchange
relative to the version 6.2 rate times 100. PPP and
US the exchange rate are
both expressed as
national currency units
per US dollar. PPP is
the number of currency
units required to buy
goods equivalent to
what can be bought with
one unit of the base
country. In the PWT,
PPP is calculated over
GDP, ie, PPP is the
national currency value
of GDP divided by the
real value of GDP in
interna tional dollars.
The international
dollar has the same
purchasing power over
total US GDP as the US
dollar in a given base
year.
y PPP-adjusted Penn World y is obtained from an
income per Tables aggregation using price
capita, measured version 6.2 parities and domestic
relative to the currency expenditures
US for consumption,
investment and
government of August
2001 vintage.
Oil Dummy for IMF (2007)
hydrocarbon-rich
countries
Dummy for net IEA (2008)
energy exporters
Price Policy reform EBRD EBRD transition
liberalisation, EBRD dummies indicators are measured
trade defined on the on a scale between 1
liberalisation, basis of EBRD and 4+ (=4.33) in steps
competition transition of one-third of a point
policy, indicators each. 1 represents no
large or little progress; 2
privatisation, indicates important
small progress; 3 is
privatisation substantial progress; 4
indicates comprehensive
progress, while 4+
indicates that
countries have reached
the standards and
performance norms of
advanced industrial
countries. Accordingly,
non-transition
countries in the sample
are evaluated at 4+.
Respective dummy
variables indicate
whether or not a
country has reached a
certain level on the
EBRD scale in the
respective policy area
within a given period.
Variable Descriptive statistics
p Mean Std. Dev Min Max
73.1749 42.6816 3.42 174.79
y 49.83053 26.57781 10.09431 100
Oil Azerbaijan, Kazakhstan, Norway,
Russia, Turkmenistan
Azerbaijan, Canada, Denmark, Kazakhstan,
Norway, Russia, Turkmenistan
Price Value Indicator (Per cent)
liberalisation,
trade Prize Trade
liberalisation, Liberalisation Liberalisation
competition
policy, 1 0.41 3.72
large 1.67 0.62 0.21
privatisation, 2 3.72 1.24
small 2.33 -- 1.03
privatisation 2.67 1.03 0.41
3 29.75 6.20
3.33 4.55 1.45
3.67 0.21 0.41
4 1.86 14.67
4.33 57.85 70.66
Variable
p
y
Oil
Price Indicator (Per cent)
liberalisation,
trade Competition Large Small
liberalisation, policy privatisation privatisation
competition
policy, 6.20 4.34 1.03
large 1.03 1.45 0.62
privatisation, 15.91 7.44 5.17
small 10.12 1.24 0.62
privatisation 3.51 1.45 0.21
9.50 13.64 3.10
-- 6.82 2.69
-- 1.65 4.13
-- 8.26 13.84
53.72 53.72 68.60
Acknowledgements
We are grateful to Roswitha King, Emilia Penkova, Volkhart Vincentz
and an anonymous referee as well as conference participants in
Regensburg, Vallendar, and Tartu for helpful comments and suggestions.
Special thanks are due to Joe Brada for editorial guidance.
REFERENCES
Balassa, B. 1964: The purchasing power parity doctrine: A
reappraisal. Journal of Political Economy 72(6): 584-596.
Barlow, D. 2010: How did structural reform influence inflation in
transition economies? Economic Systems 34(2): 198-210.
Bergin, P, Glick, R and Taylor, AM. 2006: Productivity,
tradability, and the long-run price puzzle. Journal of Monetary
Economics 53 (8): 2041-2066.
Bergstrand, JH. 1991: Structural determinants of real exchange
rates and national price levels: Some empirical evidence. American
Economic Review 81(1): 325-334.
Blanchard, O. 1997: The economics of post-communist transition.
Clarendon Press: Oxford.
Chong, Y, Jorda, O and Taylor, AM. 2012: The
Harrod-Balassa-Samuelson hypothesis: Real exchange rates and their
long-run equilibrium. International Economic Review 53(2): 609-634.
Choudhri, EU and Khan, MS. 2005: Real exchange rates in developing
countries: Are Balassa-Samuelson effects present? IMF Staff Papers
52(3): 387-409.
Cincibuch, M and Podpiera, J. 2006: Beyond Balassa-Samuelson: Real
appreciation in tradables in transition countries. Economics of
Transition 14(3): 547-573.
Clague, CK. 1985: A model of real national price levels. Southern
Economic Journal 51(4): 998-1017.
Coricelli, F and Jazbec, B. 2004: Real exchange rate dynamics in
transition economies. Structural Change and Economic Dynamics 15(1):
83-100.
de Broeck, M and Slok, T. 2006: Interpreting real exchange rate
movements in transition countries. Journal of International Economics
68(2): 368-383.
Dobrinsky, R. 2003: Convergence in per capita income levels,
productivity dynamics and real exchange rates in the EU acceding
countries. Empirica 30(3): 305-334.
Dufrenot, G and Egert, B. 2005: Real exchange rates in Central and
Eastern Europe. What scope for the underlying fundamentals? Emerging
Markets Finance and Trade 41 (2): 41-59.
Egert, B and Halpern, L. 2006: Equilibrium exchange rates in
Central and Eastern Europe: A meta-regression analysis. Journal of
Banking & Finance 30(5): 1359-1374.
Egert, B, Halpern, L and MacDonald, R. 2006: Equilibrium exchange
rates in transition economies: Taking stock of the issues. Journal of
Economic Surveys 20(2): 257-324.
Fidrmuc, J. 2009: Gravity models in integrated panels. Empirical
Economics 37(2): 435-446.
Frensch, R. 2000: Internal liberalization as a barrier to
export-led recovery in central European countries preparing for
EU-accession. Comparative Economic Studies 42(3): 31-47.
Frensch, R. 2004: Public governance as the source of quality and
variety gains from transition. Journal of Comparative Economics 32 (3):
388-408.
Frensch, R. 2006: Balassa-Samuelson, Product Differentiation and
Transition. OEI Working Paper 266, Munich, November.
Frensch, R and Schmillen, A. 2011: Can we identify
Balassa-Samuelson effects with measures of product variety? Economic
Systems 35 (1): 98-108.
Garcia-Solanes, J, Sancho-Portero, FI and Torrejon-Flores, F. 2008:
Beyond the Balassa-Samuelson effect in some new member states of the
European Union. Economic Systems 32(1): 17-32.
Halpern, L and Wyplosz, C. 2001: Economic transformation and real
exchange rates in the 2000s: The Balassa-Samuelson connection. Economic
Survey of Europe 2001(1): 227-239.
Hanousek, J and Kocenda, E. 2010: Divide and privatize: Firm
break-up and erformance. Institute for East and Southeast European
Studies. Working Paper 291, Regensburg, October.
Herrendorf, B and Valentinyi, A. 2012: Which sectors make poor
countries so unproductive? Journal of the European Economic Association
10(2): 323-341.
IEA (International Energy Agency). 2008: Key world energy
statistics. Paris.
IMF (International Monetary Fund). 2007: Guide on resource revenue
transparency. Washington, DC. Jones, C. 1994: Economic growth and the
relative price of capital. Journal of Monetary Economics 34(3): 359-382.
Lothian, JR and Taylor, MP. 2008: Real exchange rates over the past
two centuries: How important is the Harrod-Balassa-Samuelson effect?
Economic Journal 118(532): 1742-1763.
Maeso-Fernandez, F, Osbat, C and Schnatz, B. 2005: Pitfalls in
estimating equilibrium exchange rates for transition economies. Economic
Systems 29(2): 130-143.
Melitz, MJ. 2003: The impact of trade on aggregate industry
productivity and intra-industry reallocations. Econometrica 71(6):
1695-1725.
Ravallion, M. 2010: Price levels and economic growth. Making sense
of the PPP changes between ICP rounds. World Bank Policy Research.
Working Paper No. 5229, Washington, DC, March.
Samuelson, PA. 1964: Theoretical notes on trade problems. Review of
Economics and Statistics 46(2): 145-154.
Samuelson, PA. 1994: Facets of Balassa-Samuelson thirty years
later. Review off International Economics 2(3): 201-226.
Taylor, AM and Taylor, MP. 2004: The purchasing power parity
debate. Journal of Economic Perspectives 18(4): 135-158.
(1) For evidence on observation (d), based on the 1996 Penn World
Tables benchmark study, see Herrendorf and Valentinyi (2012).
(2) For a simple exposition, see, for example, Frensch (2006).
(3) The IMF's International Financial Statistics (IFS) provide
trade-weighted real effective exchange rate index series for a number of
countries that cannot be compared in levels across countries in an
economically meaningful way. Frensch (2006) performs simple OLS
regressions of yearly changes of available IFS real effective exchange
rate data for the decade between 1990 and 2000 on yearly changes of PWT
comparative prices. The estimated slope coefficient of 0.40 is
significant at the 1% level, the intercept is insignificant at the 10%
level ([R.sup.2] = 0.29; sample size = 864). Specifying country and/or
period fixed effects does not qualitatively alter the results.
Increasing the time horizon and thus eliminating nominal disturbances
even strengthens the link between the two measures. Also, differentials
between rates of change of the two measures are not systematically
related to PPP-adjusted income per capita.
(4) According to Frensch and Schmillen (2011), many empirical
studies may fail to find a significant influence of a simple BS-driven
behaviour on real exchange rate developments because of measurement
errors leading to downward-biased estimates. They test the BS hypothesis
with trade-based variety measures to differentiate between tradables and
non-tradables sector productivities that do not suffer from such
errors-in-variables and find stable and very robust BS effects.
(5) We also experimented with PMG estimations. Probably due to the
shortness of our panel, results were unstable.
(6) While our sample would ideally have included non-transition
middle-income or emerging countries, issues related to the availability
of the EBRD Transition Indicators made this infeasible. One should note,
however, that, in terms of PPP-adjusted income per capita, there is
considerable overlap between OECD and CEEC, CEEC and CIS and, because of
Turkey, even between OECD and CIS economies (cf. Figure 2).
(7) Empirically, the price reducing competition effect of trade
liberalisation is not equal across sectors: less open economies tend to
have higher investment to consumer goods price ratios than more open
economies (see, among many others, Jones, 1994).
(8) Note that this would not contradict a potentially dampening
role of price liberalisation upon inflation; for more on this, see
Barlow (2010).
(9) A 'potentially important difference is that (compared to
the 1993 and prior ICP rounds) stricter quality standards were used in
the 2005 price surveys, to assure that the ICP was obtaining prices for
internationally comparable commodities. This is important given that one
expects that lower quality goods are consumed in poorer countries,
creating a risk that (without strict standards in defining the products
to be priced) one will underestimate the cost of living in poor
countries by confusing quality differences with price differences'
(Ravallion, 2010, p. 2).
(10) With the exception of Choudhri and Khan (2005), testing the
Penn effect has in general been confined to developed countries.
(11) Recently, the Penn effect may have been attenuated: the 2005
International Comparison Program (ICP) found substantially higher PPP
rates, relative to market exchange rates, in most developing countries.
Ravallion (2010) finds that more rapidly growing economies experience
steeper increases in their price level index, while this effect has been
even stronger for initially poorer countries.
(12) Table 1 time-series results are also robust to all Table 2
specifications.
RICHARD FRENSCH [1] & ACHIM SCHMILLEN [2]
[1] Institute for East and Southeast European Studies, University
of Regensburg, Landshuter Str. 4, Regensburg 93047, Germany.
E-mail: frensch@ios-regensburg.de
[2] Institute for Employment Research, Institute for East and
Southeast European Studies, Regensburger Str. 104, Nuremberg 90478,
Germany.
E-mail: achim.schmillen@iab.de
Table 1: Comparative prices regressions
(1) (2)
OLS with OLS with
country-fixed country-fixed
effects effects
Constant 2.6273 *** (0.7260) 1.8333 ** (0.6878)
log y 0.4514 * (0.2538) 0.4514 * (0.2552)
CEEC x log y 0.4822 (0.4830) 0.2046 (0.3651)
CEEC
CIS x log y -1.6241 *** (0.4310) -0.0971 ** (0.4130)
CIS
Oil
Price liberalisation 0.2053 ** (0.0823)
Trade liberalisation 0.2633 *** (0.0789)
Competition policy 0.0946 (0.1410)
Large privatisation 0.0734 (0.0655)
Small privatisation 0.2510 ** (0.0953)
Observations (cross- 484 (41) 484 (41)
sections)
[R.sup.2] 0.33 0.51
(3)
OLS with
period-fixed
effects
Constant 2.1913 *** (0.1689)
log y 0.4760 *** (0.0219)
CEEC x log y -0.0784 (0.0715)
CEEC -0.2303 (0.2587)
CIS x log y -0.2230 (0.1768)
CIS -0.3320 (0.5121)
Oil 0.1972 *** (0.0556)
Price liberalisation 0.2366 * (0.1251)
Trade liberalisation 0.2433 ** (0.0975)
Competition policy -0.0749 (0.0876)
Large privatisation 0.0089 (0.0712)
Small privatisation 0.1935 *** (0.0530)
Observations (cross- 484 (41)
sections)
[R.sup.2] 0.92
Notes: Dependent variable: Log p; unbalanced samples of countries
with 10<[y.sup.it]<110; 1992-2004; * (**, ***) indicate
significance at 10 (5, 1)%; heteroskedasticity robust standard
errors in parentheses; results are robust to the use of
bootstrapped standard errors and bias correction (200
replications), except for the log y coefficients in columns 1 and 2
and the small privatisation coefficient in column 2, which become
significant at the 1% level and the price liberalisation
coefficient in column 3, which becomes significant at the 5% level.
Table 2: Comparative prices regressions with period-fixed effects
(4) (5)
Without Armenia, Extended sample:
Azerbaijan and 1989-2004
Kyrgyzstan
Constant 2.1735 *** (0.1680) 2.2902 *** (0.1310)
Log y 0.4841 *** (0.0214) 0.4736 *** (0.0205)
CEEC x log y -0.0610 (0.0714) -0.0200 (0.0625)
CEEC -0.2818 (0.2573) -0.4152 * (0.2271)
CIS x log y -0.2083 (0.1957) -0.1204 (0.1866)
CIS -0.2864 (0.5438) -0.6718 (0.5173)
Oil 0.1240 *** (0.0459) 0.1844 *** (0.0534)
Price liberalisation 0.2835 ** (0.1126) 0.1629 (0.1046)
Trade liberalisation 0.2420 ** (0.1025) 0.2022 ** (0.0847)
Competition policy -0.1979 ** (0.0868) -0.1904 ** (0.0829)
Large privatisation 0.0565 (0.0738) 0.0314 (0.0678)
Small privatisation 0.2088 *** (0.0550) 0.1752 *** (0.0498)
Observations (cross- 472 (38) 568 (41)
sections)
[R.sup.2] 0.92 0.90
(6) (7)
Oil dummy for all Reform dummies at
net energy 1/3 of cumulative
exporters according distribution
to IEA (2008)
Constant 2.2040 *** (0.1709) 2.3086 *** (0.1821)
Log y 0.4752 *** (0.0225) 0.4797 *** (0.0227)
CEEC x log y -0.0740 (0.0722) -0.0425 (0.0589)
CEEC -0.2405 (0.2609) -0.3683 * (0.2123)
CIS x log y -0.1747 (0.1902) -0.2442 (0.1900)
CIS -0.4216 (0.5224) -0.3741 (0.5178)
Oil 0.0942 ** (0.0396) 0.1700 *** (0.575)
Price liberalisation 0.2416 * (0.1252) 0.2059 (0.1477)
Trade liberalisation 0.2291 ** (0.0979) 0.0687 (0.1793)
Competition policy -0.0993 (0.0877) -0.0846 (0.0748)
Large privatisation 0.0201 (0.0711) -0.0735 (0.1430)
Small privatisation 0.2013 *** (0.0527) 0.3558 *** (0.0739)
Observations (cross- 484(41) 484 (41)
sections)
[R.sup.2] 0.91 0.91
(8)
Reform dummies at
2/3 of cumulative
distribution
Constant 2.0698 *** (0.1703)
Log y 0.4744 *** (0.0216)
CEEC x log y -0.1601 ** (0.0713)
CEEC 0.1018 (0.2576)
CIS x log y -0.2542 (0.1880)
CIS -0.1515 (0.5138)
Oil 0.2109 *** (0.0531)
Price liberalisation 0.2222 * (0.1287)
Trade liberalisation 0.2378 ** (0.0981)
Competition policy 0.1149 ** (0.0522)
Large privatisation -0.0273 (0.0707)
Small privatisation 0.1812 *** (0.0535)
Observations (cross- 484 (41)
sections)
[R.sup.2] 0.92
Notes: Dependent variable: log p; unbalanced samples of countries
with 10 < [y.sub.jt] < 110; 1992-2004 (1989-2004 in column 4); *
(**, ***) indicate significance at 10 (5, 1)%; heteroskedasticity
robust standard errors in parentheses; results are qualitatively
robust to the use of bootstrapped standard errors and bias
correction (200 replications).
