Tradability of output and the current account in Europe

Abstract

This paper investigates empirically the relationship between specialisation in the production of tradable output and the current account balance. According to the ‘tradability hypothesis’ that is examined, countries that specialise in highly tradable sectors tend to run current account surpluses while countries with dominant non-tradable sectors risk running current account deficits. In order to test this hypothesis empirically we develop a value-added based tradability index which captures the tradability of a country’s output. Applied to a large sample of European countries, our empirical model provides strong evidence in favour of the tradability hypothesis. The main policy implication is that the anxieties about ‘de-industrialisation’ in large parts of Europe seem justified with a view to growing external imbalances.

Appendices

Appendix 1: List of countries

Country code	Country	Country category
AL	Albania	Emerging
AM	Armenia	Emerging
AT	Austria	Developed
AZ	Azerbaijan	Emerging
BY	Belarus	Emerging
BE	Belgium	Developed
BA	Bosnia and Herzegovina	Emerging
BG	Bulgaria	Emerging
HR	Croatia	Emerging
CY	Cyprus	Emerging
CZ	Czech Republic	Developed
DK	Denmark	Developed
EE	Estonia	Developed
FI	Finland	Developed
FR	France	Developed
GE	Georgia	Emerging
DE	Germany	Developed
EL	Greece	Developed
HU	Hungary	Developed
IS	Iceland	Developed
IE	Ireland	Developed
IT	Italy	Developed
KZ	Kazakhstan	Emerging
LV	Latvia	Developed
LT	Lithuania	Emerging
LU	Luxembourg	Developed
MK	Macedonia	Emerging
MT	Malta	Emerging
MD	Moldova	Emerging
ME	Montenegro	Emerging
NL	Netherlands	Developed
NO	Norway	Developed
PL	Poland	Emerging
PT	Portugal	Developed
RO	Romania	Emerging
RU	Russia	Emerging
RS	Serbia	Emerging
SK	Slovakia	Developed
SI	Slovenia	Developed
ES	Spain	Developed
SE	Sweden	Developed
CH	Switzerland	Developed
TR	Turkey	Emerging
UA	Ukraine	Emerging
UK	United Kingdom	Developed
XK	Kosovo	Emerging

1. The distinction between ‘Developed’ and ‘Emerging’ mirrors the categorisation of European countries as ‘Advanced’ and ‘Emerging and Developing’ by the IMF as of April 2014

Appendix 2: List of sectors for the calculation of the tradability index

Number	Sector	NACE Rev. 1	NACE Rev. 2
1	Agriculture, hunting and forestry + Fishing	A + B	A
2	Mining and quarrying	C	B
3	Manufacturing	D	C
4	Electricity, gas and water supply	E	D + E
5	Construction	F	F
6	Wholesale, retail trade, repair of motor vehicles etc.	G	G
7	Hotels and restaurants	H	I
8	Transport, storage + Communication	I	H + J
9	Financial intermediation	J	K
10	Real estate, renting and business activities	K	L + M + N
11	Public administration, defence, compuls.soc.security	L	O
12	Education	M	P
13	Health and social work	N	Q
14	Other community, social and personal services + Private households with employed persons	O + P	R + S + T

Appendix 3: Methodology for calculating value added exports

Deriving the tradability score requires the calculation of the value added exports (VAX) at the industry-country level. This appendix illustrates the basic input-output methodology to calculate the VAX, including a 3-country, 2-sector example.

Following the trade in value added concept in Johnson and Noguera (2012) and the expositions in Stehrer (2012), three components are required to calculate the value added exports. For any reporting country r, these components are the value added requirements per unit of gross output, v^r; the Leontief inverse of the global input-output matrix, L; and the final demand vector, f^J where J indicates that the vector includes demand from every country j ∈J. Both vectors as well as the Leontief inverse have an industry dimension i.

Country r’s value added coefficients are defined as ${v}^r=\frac{{value\ added}_r}{{gross\ output}_r}$. The value added coefficients are arranged in a diagonal matrix of dimension 1435 × 1 (41 countries × 35 industries). This matrix contains the value added coefficients of country r for all industries along the diagonals. The remaining entries of the matrix are zero because the interest here is with the value added created in country r.

The second element is the Leontief inverse of the global input-output matrix, L = (I − A)⁻¹ where A denotes the coefficient matrix. In the World Input-Output Database (WIOD) the coefficient matrix (and hence the Leontief matrix) is of dimension 1435 × 1435 which contains the technological input coefficients of country r in the diagonal elements and the technological input coefficients of country r’s imports (from a column perspective) and exports (from a row perspective) in the off-diagonal elements.

The final building block is the global final demand vector. This vector is also industry-specific and is of dimension 1435 × 1. Importantly, the final demand comes from different countries (i.e. the origin of the final demand). As usual, each of the rows is associated with one source country of the goods and services that are finally demand. Therefore, the full final demand vector, f^J, in the 3-country, 2-sector case, with m (for manufacturing) and s (for services) indicating the sector (or industry), has the following form:

$$ {f}_i^J=\left(\begin{array}{c}{f}_m^{r,r}+{f}_m^{r,2}+{f}_m^{r,3}\\ {}{f}_s^{r,r}+{f}_s^{r,2}+{f}_s^{r,3}\\ {}{f}_m^{2,r}+{f}_m^{2,2}+{f}_m^{2,3}\\ {}{f}_s^{2,r}+{f}_s^{2,2}+{f}_s^{2,3}\\ {}{f}_m^{3,r}+{f}_m^{3,2}+{f}_m^{3,3}\\ {}{f}_s^{3,r}+{f}_s^{3,2}+{f}_s^{3,3}\end{array}\right) $$

where the subscript J indicates that the vector comprises the consumption of all countries j ∈ J. The typical element of this vector contains the final demand from all possible countries. For example, the element ${f}_m^{r,3}$ captures the value of final goods and services that country 3 demands from the manufacturing sector in country r. Since the idea of value added exports is that they comprise only value added that is created in one country but absorbed in another, the final demand from country r itself needs to be eliminated for the calculation of country r’s VAX. An adjusted final demand vector, f^j ≠ r, is used in which country r’s final demand (i.e. the first column in the above matrix) is set to zero. Country r’s value added exports can then be calculated as

$$ {VAX}^{r,\ast }={\boldsymbol{\upsilon}}^{\boldsymbol{r}}\cdotp \boldsymbol{L}\cdotp {f}^{j\ne r} $$

(3)

where VAX^{r, ∗} are the sector-specific value added exports of country r to all partner countries.

To illustrate this, the detailed matrices in the 3 country, 2-sector case, where country r is still the model country are shown. Equation (3) then has the following form:

$$ \left(\begin{array}{c}{VAX}_{m,\ast}^{r,\ast}\\ {}{VAX}_{s,\ast}^{r,\ast}\\ {}0\\ {}0\\ {}0\\ {}0\end{array}\right)=\left(\begin{array}{cccccc}\ {\nu}_m^r& 0& 0& 0& 0& 0\\ {}0& {\nu}_s^r& 0& 0& 0& 0\\ {}0& 0& 0& 0& 0& 0\\ {}0& 0& 0& 0& 0& 0\\ {}0& 0& 0& 0& 0& 0\\ {}0& 0& 0& 0& 0& 0\end{array}\right)\cdotp \left(\begin{array}{cccccc}{l}_{m,m}^{r,r}& {l}_{m,s}^{r,r}& {l}_{m,m}^{r,2}& {l}_{m,s}^{r,2}& {l}_{m,m}^{r,3}& {l}_{m,s}^{r,3}\\ {}{l}_{s,m}^{r,r}& {l}_{s,s}^{r,r}& {l}_{s,m}^{r,2}& {l}_{s,s}^{r,2}& {l}_{s,m}^{r,3}& {l}_{s,s}^{r,3}\\ {}{l}_{m,m}^{2,r}& {l}_{m,s}^{2,r}& {l}_{m,m}^{2,2}& {l}_{m,s}^{2,2}& {l}_{m,m}^{2,3}& {l}_{m,s}^{2,3}\\ {}{l}_{s,m}^{2,r}& {l}_{s,s}^{2,r}& {l}_{s,m}^{2,2}& {l}_{s,s}^{2,2}& {l}_{s,m}^{2,3}& {l}_{s,s}^{2,3}\\ {}{l}_{m,m}^{3,r}& {l}_{m,s}^{3,r}& {l}_{m,m}^{3,2}& {l}_{m,s}^{3,2}& {l}_{m,m}^{3,3}& {l}_{m,s}^{3,3}\\ {}{l}_{s,m}^{3,r}& {l}_{s,s}^{3,r}& {l}_{s,m}^{3,2}& {l}_{s,s}^{3,2}& {l}_{s,m}^{3,3}& {s}_{s,s}^{r,3}\end{array}\right)\cdotp \left(\begin{array}{c}0+{f}_m^{r,2}+{f}_m^{r,3}\\ {}\ 0+{f}_s^{r,2}+{f}_s^{r,3}\ \\ {}\begin{array}{c}0+{f}_m^{2,2}+{f}_m^{2,3}\\ {}0+{f}_s^{2,2}+{f}_s^{2,3}\\ {}0+{f}_m^{3,2}+{f}_m^{3,3}\\ {}0+{f}_s^{3,2}+{f}_s^{3,3}\end{array}\end{array}\right) $$

The coefficients in the Leontief matrix represent the total direct and indirect input requirements of any country in order to produce one dollar worth of output for final demand. For example, the coefficient ${l}_{m,s}^{r,r}$ indicates the input requirement of country r’s services sector from country r’s manufacturing sector for producing one unit of output. Likewise, the coefficient ${l}_{m,m}^{r,3}$ indicates country r’s input requirement in the manufacturing sector supplied by country 3’s manufacturing sector.

The resulting elements in this example, ${VAX}_{m,\ast}^{r,\ast }$ and ${VAX}_{s,\ast}^{r,\ast }$, are the total value added exports of country r that originate from the manufacturing and services sector respectively, to all other sectors of all partner countries.

The VAX are not only calculated for country r but for all 40 countries plus the rest of the world.

The final step needed to arrive at the global industry-level VAX is to sum up the VAX of all countries for each individual sector i. Dividing the global industry-specific VAX by the corresponding industry-specific value added yields the tradability score by sector. In that process, also the more detailed industry structure available in the WIOD database (35 industries) is aggregated to the 14 broad sectors on which the tradability index is based.

Appendix 4: The effects of a structural shock in an inter-temporal current account model

This appendix provides simulation-based insights into the effects of a structural shock towards non-tradable goods in a very simple two-period inter-temporal model of the current account that features a tradable and a non-tradable sector (Obstfeld and Rogoff 1996). As will be shown, the tradability hypothesis that is tested empirically can be derived from such as model under reasonable parameter settings. The first part of this Appendix presents the basic model, while the second section describes the actual outcomes of the structural shock of interest.

1.1 The inter-temporal current account model

Consumption

There are only two periods. The representative consumer derives her utility from consumption in periods t = 1 and t = 2. Preferences are characterised by a constant inter-temporal elasticity of substitution, $\frac{1}{\sigma }$, resulting in a utility function with constant relative risk aversion of the form²⁸

$$ U=\frac{C_1^{1-\sigma }-1}{1-\sigma }+\beta \cdotp \frac{C_2^{1-\sigma }-1}{1-\sigma } $$

(4)

where C_t is the instantaneous consumption in period t and β is the discount factor indicating the consumer’s patience with regard to postponing consumption. In each period the consumer splits consumption between the consumption of the tradable good $\left({C}_t^T\right)$and the consumption of the non-tradable good $\left({C}_t^N\right)$ assigning a constant fraction to each of the two goods. This gives rise to Cobb-Douglas style preferences

$$ {C}_t={\left({C}_t^T\right)}^{\gamma}\cdotp {\left({C}_t^N\right)}^{1-\gamma } $$

(5)

where γ is the consumption share of the tradable good.²⁹

Cost minimisation yields the demand functions for the two types of goods

$$ {C}_t^T=\gamma \cdotp \left(\frac{1}{P_t^T}\right)\cdotp {P}_t\cdotp {C}_t $$

(6)

$$ {C}_t^N=\left(1-\gamma \right)\cdotp \left(\frac{1}{P_t^N}\right)\cdotp {P}_t\cdotp {C}_t. $$

(7)

Given that P_t is the numeraire, the price of the non-tradable good, ${P}_t^N$, is determined by the relative consumption of the two goods.

Since the expenditure share of the two goods are fixed, the price index takes the simple form

$$ {P}_t={\left(\frac{1}{\gamma}\right)}^{\gamma}\cdotp {\left(\frac{1}{1-\gamma}\right)}^{1-\gamma}\cdotp {\left({P}_t^N\right)}^{1-\gamma } $$

(8)

Once the intra-temporal choices are made, the consumer chooses the optimal consumption path over time. For the postponement of consumption until period 2 of any of her income, the consumer is compensated with the interest rate r which is assumed to be determined on international capital markets and therefore exogenous. The consumer maximises lifetime utility in (4) under the intertemporal budget constraint

$$ {C}_1\cdotp {P}_1+\frac{C_2\cdotp {P}_2}{1+r}={Y}_1^T+{P}_1^N\cdotp {Y}_1^N+\frac{Y_2^T+{P}_2^N\cdotp {Y}_2^N}{1+r} $$

(9)

which simply states that the sum of consumption in periods 1 and 2 must equal the sum of the (exogenous) output in the two periods. This maximisation problem leads to the usual Euler equation

$$ {C}_1^{-\sigma }=\beta \cdotp \left(1+r\right)\cdotp \left(\frac{P_1}{P_2}\right)\cdotp {C}_2^{-\sigma } $$

(10)

with σ > 0.

The more patient the consumer is, i.e. the higher β, the more is she prepared to postpone consumption until period 2. Likewise, the higher the interest rate, the greater is the resulting income effect and the more consumption is shifted to period 2. Finally, a lower relative inter-temporal price level, $\frac{P_1}{P_2}$, (i.e. a decline in the price index over time) will induce the consumer to shift consumption towards period 2.

1.1.1 The current account

The current account in period 1 is simply the difference between the consumer’s endowments with tradable goods and the consumption of tradable goods:

$$ {CA}_1={Y}_1^T-{C}_1^T $$

(11)

Taking into account that in each period the consumption of the non-tradable good equals the endowment, the intertemporal budget constraint in (12) simplifies to

$$ {C}_1^T+\frac{C_2^T}{1+r}={Y}_1^T+\frac{Y_2^T}{1+r} $$

(12)

Combining the equation for the current account with the demand function for the tradable good (6), the Euler eq. (10) and the simplified version of the intertemporal budget constraint (12) yields the following expression for the current account equation in period 1:

$$ {CA}_1={Y}_1^T-\frac{Y_1^T+\frac{Y_2^T}{1+r}}{1+{\beta}^{\frac{1}{\sigma }}\cdotp {\left(1+r\right)}^{\frac{1-\sigma }{\sigma }}\cdotp {\left(\frac{P_2}{P_1}\right)}^{\frac{\sigma -1}{\sigma }}}. $$

(13)

The demand function for the non-tradable good (7), together with the price index (8) and the Euler eq. (10) yields an expression for the relative demand for the non-tradable good in the two periods which depends on the relative inter-temporal prices and the intertemporal elasticity of substitution (IES). Since in the case of the non-tradable good consumption equals endowment in each period one obtains:

$$ \frac{Y_2^N}{Y_1^N}={\left(\frac{P_1}{P_2}\right)}^{\frac{1-\sigma }{\sigma }+\frac{1}{1-\gamma }}\ \mathrm{or}\ \mathrm{equivalently}\ \frac{P_1}{P_2}={\left(\frac{Y_2^N}{Y_1^N}\right)}^{\frac{\left(1-\gamma \right)\cdotp \sigma }{1-\gamma +\gamma \sigma}} $$

(14)

This equation describes the development of the price index in relation to the development of the supply of the non-tradable good over time. Since 0 < γ < 1 and σ > 0, an increase in the supply of the non-tradable good over time, i.e. an increase in${Y}_2^N$, leads to a decrease in the price index in period 2 (P₂) and consequently an increase in the relative price index $\frac{P_1}{P_2}$.

Inserting (14) into (13) yields the formulation of the current account equation as stated in the main text:

$$ {CA}_1={Y}_1^T-\frac{Y_1^T+\frac{Y_2^T}{1+r}}{1+{\beta}^{\frac{1}{\sigma }}\cdotp {\left(1+r\right)}^{\frac{1-\sigma }{\sigma }}\cdotp {\left(\frac{Y_2^N}{Y_1^N}\right)}^{\frac{\left(1-\gamma \right)\cdotp \left(1-\sigma \right)}{1-\gamma +\gamma \sigma}}}. $$

(15)

This expression states the current account position as a function of the endowments in the tradable and the non-tradable good, the inter-temporal elasticity of substitution, the consumption expenditure share of the tradable-good, γ, (which is assumed to be constant), the interest rate and the discount factor (β). This equation is used to examine the effects of an endowment-neutral structural shock on the current account and the tradability index.

1.2 The effects of a structural shocks towards non-tradables on the current account balance

As mentioned in the main text, the tradability hypothesis is compatible with the inter-temporal current account model’s predictions for the current account position resulting from a non-tradables shock. More precisely, the non-tradables shock is an endowment-neutral structural shock in period 2 that consists of an increase in the economy’s endowment with the non-tradable good Y^Nand a simultaneous decrease in its endowment with the tradable good Y^Tby the same amount. This allows highlighting the impact of the structural effect by switching off the wealth effect which would arise if only Y^Tor Y^N were to change. This endowment-neutral structural shock corresponds to a decline in the tradability index (TI) in the empirical analysis.

The main adjustment mechanism in the inter-temporal endowment economy is the consumption decisions made by the welfare maximising representative consumer.^. The effect of the adjustments of the consumption plans (induced by the structural shock) on the current account works via two channels. The first channel is that the consumer postpones consumption into period 2 because she wants to consume the tradable and the non-tradable good in fixed proportions, governed by the expenditure share of the tradable good, γ (structural consumption adjustment effect). The second channel is the decline in the price index over time which means that the consumption-based interest rate increases (income effect). Hence, under reasonable parameter settings, the structural shock that is of interest here will cause a deterioration of the current account.³⁰

We rely on simulations to show the relationship between the tradability of output and the current account for plausible values of the expenditure shares γ and 1 − γ, the initial endowments ${Y}_1^T$ and ${Y}_2^N$ and the inter-temporal elasticity of substitution σ.

Let the initial situation be characterised by a balanced current account with the parameterisation shown in Table 6.

Table 6 Parameters for the simulation of the structural shock towards non-tradables

Table 7 shows the effect on the tradability index, $\raisebox{1ex}{${Y}_t^T\cdotp {P}_t^T$}\!\left/ \!\raisebox{-1ex}{${Y}_t\cdotp {P}_t$}\right.$, and the current account of a structural shock in period 2 consisting of a decline in ${Y}_2^T$ by 20 unit and an increase in ${Y}_2^N$ of equal size. In the initial situation the current account is balanced in both periods. Given that the expenditure share on the tradables is 0.2 and the tradable good’s share in the economy’s total endowment is also 0.2, the prices of the tradable and the non-tradable goods are both equal to 1 in both periods and the price index is equal to 1.65 in both periods.

Table 7 Effects of a structural shock on the current account and the tradability index

The consequence of the structural shock in period 2 is a current account surplus in period 1 and a corresponding deficit in period 2. The surplus in period 1 is explained by two effects.

First, there is a structural consumption adjustment effect. This structural adjustment effect stems from the fact that the consumer maximises utility when she consumes the tradable and the non-tradable good exactly in the proportions dictated by the expenditure share. In the above example this is a ratio of 1 to 4. The structural shock that hits in period 2 reduces the tradables to non-tradables ratio in terms of endowments to a much lower level. This creates an incentive to shift some of the tradable good consumption from period 1 to period 2. The extra utility gained in period 2 (ΔC₂) from this shift exceeds the loss in aggregate consumption in period 1 (ΔC₁).

Second, there is an income effect which stems from the fact that the price index is declining over time because the endowment of the non-tradable good increases in period 2. The declining price index represents an increase in the consumption-based interest rate and is therefore an incentive for the consumer to postpone consumption until period 2. This second effect also contributes to the current account surplus in the first period. Due to the terminal condition, the resulting current account position in period 2 is a corresponding deficit.

The magnitude of both effects is determined by the extent to which consumption is postponed which in turn depends on the IES. If the IES is large, relatively more consumption will be shifted into period 2 as the consumer is relatively more prepared to accept fluctuations in the consumption path over time. Note that these results are not due to a wealth effect, which is neutralised by simultaneously increasing the endowment of the non-tradable good and reducing the endowment with the tradable good by the same amount (i.e. $\mathit{\Delta }{Y}_2^N=-\mathit{\Delta }{Y}_2^T$ with $\mathit{\Delta }{Y}_2^N>0$).

Hence, with the parameter constellation as in Table 6 the result of the structural shock towards non-tradables at the time when that shock materialises, i.e. period 2, is that the TI declines as does the current account balance. It is also clear that in period 2, a country that suffers from such a structural shock will have a lower TI and a worse current account position than a country that does not experience such a shock (given that both countries start with the same initial conditions). Figure 9 illustrates the evolution of the current account balance, the tradability index and the price index for various values of the inter-temporal elasticity of substitution (IES).

Fig. 9

Current account, TI and relative prices after the structural shock (${Y}_1^T=100$; ${Y}_1^N=400$). Note: Parameters as in Table 6 except for the inter-temporal elasticity of substitution (IES); structural shock in t₂: $\Delta {Y}_1^T=-20;\Delta {Y}_1^N=+20$

Appendix 5: Additional regression results

1.1 Results by country groups

Estimating the cross-section model in eq. (1) for individual country groups entails the problem that the number of observations is getting very small. Nevertheless we perform this analysis as one of several robustness checks.

The model is re-estimated separately for developed and emerging European countries as well as the EU Member States, euro area members and the Central, Eastern and South Eastern European (CESEE) region. We further report results for the entire sample excluding Azerbaijan and Montenegro which are outliers regarding the current account balance as well as the entire sample excluding Ireland, Lithuania and Malta for which the fit between the country-level tradability scores and the global scores is somewhat weaker.³¹

As shown in Table 8 the results – and in particular the result regarding the tradability index – hold throughout all sub-samples. As expected the tradability of output matters more for the external balance in the case of emerging countries than for developed countries, though this results is also influenced by the oil exporters within this group. Potential explanations for this finding are that the components of the current account which are not (or only indirectly) linked to the tradability of output, such as payments of factor incomes, play a larger role in developed countries and that industrial countries only require a smaller industrial base which is increasingly interlinked with services and performs an important carrier function (see Stöllinger et al. 2013) by absorbing a large amount of services which are exported only indirectly via manufactures. An interesting result is also that the government balance turns out to be positive and highly significant for the industrialised European countries and the EU Member States but not the emerging markets. Hence, for the former, the twin deficit hypothesis seems to have some relevance. Some of the other control variables which are found to be statistically significant in the specification using the full sample are also very robust across the various sub-samples such as the net foreign asset position and the relative GDP per capita. For others no statistically significant effects can be established though it is hard to tell whether this is due to the small number of observations or the particularities of the respective country group.

Finally specification (7) and specification (8) show that the cross-section results are not driven by outliers.

Table 9 provides the same robustness check for the panel regression model. Also in this case the tradability index is reported to have a positive and statistically significant coefficient for each country group. The relative size of the coefficients of the TI across country groups indicates that the tradability of output a country produces matters more for emerging Europe than for the developed European economies, thereby confirming the outcome of the cross-section regressions.

There are also a few differences across the country groups with regard to some of the control variables, notable the government balance where for the developed countries the twin deficit hypothesis is confirmed whereas for the emerging European countries the coefficient of the government balance has a negative sign.

Table 8 Cross-section regression, various subsamples

Table 9 Panel regression, various subsamples

1.2 The role of the real exchange rate

Some of the specifications in the main text included the real effective exchange rate based on unit labour costs (REER_ulc)._Here the complex issue of the real exchange rate is discussed in some more detail and additional regression results using alternative measures of the exchange rate are reported.

Relative prices (respectively changes thereof), that is the bilateral real exchange rate (respectively changes thereof) influence the current account positions via exports and imports. As long as the Marshall-Learner condition is fulfilled, an increase in the relative price level worsens the current account balance by hampering exports and facilitating imports.

In our context, the issue is complicated by the fact that apart from this direct channel, relative prices may affect the current account also indirectly via its impact on countries’ specialisation patterns. Let’s assume a decline in the domestic price level (real exchange rate) caused, for example, by some change in the underlying labour market institutions such as an agreement between employers’ associations and the trade unions on wage moderation. The wage moderation policy tends to depress the overall price level relative to the trading partners (real depreciation). This is because wage moderation implies that wages are progressing at a slower pace than productivity resulting in a fall of production costs (i.e. the wage mark-up over marginal labour productivity declines). A first consequence will be that the price of non-tradables declines. The price of tradables remains constant due to the law of one price.³² This will cause a shift of domestic resources from the non-tradable sector to the tradable sector because of increased profit opportunities in the latter until the relative price between non-tradables and tradables has adjusted accordingly. Resources will be drawn into the tradable sector and the domestic price level P (defined as P^N/P^T) will adjust to equalise the marginal revenue product in the two sectors. So the key results will be an expansion of the tradable sector and a depreciation of the real exchange change. Since wage moderation will either reduce income or keep it unchanged, there will be no growth effect as in the case of a (positive) productivity shock that will cause households to shift consumption forward. For this reason, wage moderation should also lead to an improvement of the current account position (see e.g. Berthou and Gaulier 2013).

Wage moderation is one scenario in which changes in the real exchange rate will affect the composition of output (and hence our tradability index). However, the causality may go in both directions. Given that, depending on the ultimate (exogenous) shock, changes in the production structure drive the development of domestic prices and the real exchange rate or vice versa, the simplest way to integrate exchange rate developments into the empirical analysis is to include it as an additional explanatory variable into the regression model. The expectation is that a higher relative price level, respectively an appreciation of the real exchange rate, worsens the current account. The main interest, however, is not with the sign of the real exchange rate (or changes thereof) but how its inclusion affects the coefficient of the tradability index. More precisely, if the specialisation patterns and resulting production structures were predominantly the result of relative prices, the real exchange rate should pick up this effect rendering the TI variable superfluous. In other words the inclusion of a measure for the real exchange rate acts as a robustness check ruling out the possibility that the tradability index only captures changes in relative prices.

The above reasoning presumed that the price of one law holds. If the law of one price were to hold for tradable goods, changes in the real exchange rate would be entirely due to the evolution of the price of non-tradables. While a convenient assumption, the law of one price is not supported by empirical research. By decomposing real exchange rate movements, Engel (1999) and Betts and Kehoe (2006) show for the US that these movements are predominantly explained by deviations of relative prices of tradable goods and not change in the relative price of non-tradables. Drozd and Nosal (2009) confirm this finding for a sample of 21 countries assigning on average only a third of real exchange rate movements to the non-tradable sector. The results in Burstein et al. (2006) suggest a somewhat greater role for the price of non-tradables in a sample of OECD countries, giving both sectors approximately equal importance for determining real exchange rate movements. Drozd and Nosal (2010) also find a greater role for price developments in the non-tradable sector in European countries than in other countries which they assign to the fact that nominal exchange rates are fixed in the euro area. A similar argument for this phenomenon is found in Ruscher and Wolff (2009). Irrespective of the exact contribution of international price differences in tradables on the one hand and the non-tradable sector on the other hand, for the empirical analysis it is useful to keep in mind that fluctuations in the exchange rate may be caused by both.

To include an overall measure for the real exchange rate we use the price level of consumption from the Penn World Tables (version 8.1) as well as the real effective exchange rate based on unit labour costs which is used in the main text. The latter indicator is also a common measure for countries’ international competitiveness. If prices were equal to labour productivity, the unit labour cost based real effective exchange rate would equal the nominal exchange rate. A third indicator used is the measure of undervaluation or overvaluation of the exchange rate (over_eval) suggested by Dollar (1992). The measure is based on the price level of consumption and exploits the empirical regularity that the price level is generally higher in countries with higher per capita income. We follow the approach by Rodrik (2008) in estimating the expected real effective exchange rate (or relative price level) by regressing the log of the price level of consumption on the log of GDP per capita controlling for time fixed effects. The difference between the actual price level and the predicted price level is the degree to which the real exchange rate is overvalued (over_eval). A value greater than 0 indicates that a country’s real exchange rate is overvalued, values smaller than 0 indicate an undervalued real exchange rate.

Given the link between a higher real effective exchange rate and the current account described above, a negative sign for the coefficient of the exchange rate measure is to be expected.

We test for the impact of these three exchange rate measures (see Table 10).

The major insight from these specifications using alternative real exchange rate measures is that the magnitude of the coefficient of the tradability index remains largely unaffected by the inclusion of the real exchange rate measures. Among the exchange rate measures, only the changes in the unit labour cost based REER (see specification 4) turn out to be statistically significant, though only marginally.³³ As expected the sign is negative implying that a real appreciation makes it more difficult to not run a current account deficit. Our conclusion from this is it is very unlikely that the positive relationship between the TI and the current account is due to a spurious correlation driven by changes in relative prices.

Table 10 The role of the real exchange rate (cross-section regression)

1.3 Tradability of output and the trade balance

As another robustness check we re-run the regression model in eq. (1), replacing the current account balance with the trade balance as the dependent variable. The expectation is that the relationship between the TI and the trade balance is even stronger than between the TI. This is because the tradability of output should affects the current account and the trade account in the same manner, only that the former also contains other elements (notably net income and transfers) which may dilute the relationship. The results in Table 11 fully confirm this expectation.

Table 11 Tradability of output and the trade balance (cross-section regression)

These results are informative as they indicate that the income balance and the transfer balance are highly relevant elements for the position of the current account. At the same, the tradability of output which affect the current account via the trade balance still seems to be a key determinant for the position of the current account and not only to the trade balance. In other words, net income received and net transfers, on average, are still insufficient to compensate for an increasing specialisation in the production of non-tradable output.

As shown in Table 12, the tradability hypothesis is equally confirmed when applied to the trade in the panel regression model.

Table 12 Tradability of output and the trade balance (panel regression)

Appendix 6: Unit root tests

This appendix provides additional information on the unit root tests.

While the time series dimension of our panel data covers only 20 years, we check all the variables for the presence of a unit root. This is advisable since in case of a unit root present in any (or several) of the time series, the regression results may be spurious (see Granger and Newbold 1974). Therefore we perform standard panel unit root tests opting for the methods suggested by Im et al. (2003) and the Fisher-type tests (see Maddala and Wu 1999). Both the Im-Pesaran-Shin (IPS) test and the Fisher-type tests have as the null-hypothesis that all panels contain a unit root against the alternative that some panels are stationary. For the IPS test we let the lag structure be determined by the Akaike information criterion (AIC), for the Fisher-type tests we include a one-period lags, except for the government balance in the test with time trend where two lags are included (given that the number of lags suggested by the AIC in the IPS test is closer to two).

We rely mainly on the IPS tests and use the Fisher tests as a robustness check only. According to the IPS tests for the main series, i.e. the current account series and the TI series, the null-hypothesis can be rejected, although for the TI in the test without a time trend only at the 10% level (see Table 13). Since the Fisher-type tests in this case clearly reject the null-hypothesis and the TI series for most countries do seem to have a time trend we consider the unit root tests as satisfied.

With regard to the control variables the tests signal the presence of a unit root in the case of the net foreign assets, the dependency ratio and the domestic credit. For this reason we let these variables enter the panel regression model in first differences. The differenced variables clearly pass the unit root tests (see Table 14).

Table 13 Panel unit root tests, excluding time trend

Table 14 Panel unit root tests for first differenced variables

Appendix 7: Panel regression results in first differences

In Table 15 the results from the regression model in eq. (2) using first differences of all variables are presented. This should avoid the possibility that results may be invalid due to the data (or parts thereof) being integrated of order one. Certainly, the differenced version of the panel model should only be given a short term interpretation. We use this specification mainly to show that the confirmation of the tradability hypothesis in the data is not spurious.

Table 15 Panel regression results in first differences

Regarding the choice of the appropriate estimator, a Hausman test rejects the appropriateness of the random effects model. The subsequent F-test suggests that the inclusion of the country fixed effects is not necessary. Hence, in contrast to the panel model in levels, when estimating the model in first differences, the pooled model in specification 1 (which in this case includes time fixed effects) can be considered as the appropriate model. However, also here the result is not sensitive to the choice of the panel estimator.