On the Importance of Labour Productivity Growth: Portugal vs. Ireland

Transcription

1 UNIVERSITE DE LAUSANNE ECOLE DES HAUTES ETUDES COMMERCIALES Macroeconomic Modelling On the Importance of Labour Productivity Growth: Portugal vs. Ireland Cátia Felisberto July 2003 Professor: Jean-Christian Lambelet Assistant: Alexander Mihailov

2 Abstract The purpose of this paper is to analyse the importance of labour productivity and relate it with economic growth, living standards and inflation. Two particular countries are considered: Portugal and Ireland. In a first part, a growth accounting analysis is carried out, for both countries, in order to give us some insight about the behaviour of these economies and about the determinants of their economic growth. In a second part, a vector autoregressive analysis is performed, which will allow the study of the relationships between labour productivity and economic growth, living standards and inflation. We conclude that total factor productivity has been a major engine of growth over the 1990s in Ireland. Factors such as increased labour productivity and efficiency, use of labour in an intensive way and foreign investment (whose repercussions are spread to the whole economy and through time) have largely been contributing to Irish economic growth. On the other hand, in Portugal, labour and capital as major production factors account for a large portion of the explained growth. Portugal needs to increase his average level of education and professional formation to achieve a higher labour productivity. Another problem seems to be the type of investment made in Portugal which is not very fruitful in the long run. The above conclusions are corroborated by all the estimation methods applied. 1

3 I. Introduction Many studies have been dedicated to the importance of labour productivity growth. The reason is that the rate of productivity growth can have an enormous effect on real output and living standards (Steindel and Stiroh, 2001). The aim of this paper is to analyse the importance of labour productivity in the case of Portugal and Ireland. The choice of these two countries was mainly related with the fact that during many decades both countries have had similar economic growth rates and suddenly, in the 90s, Ireland started to show an annual growth rate of GDP (Gross Domestic Product) higher than the one presented by Portugal. The choice of Portugal and Ireland was also motivated by the fact that both countries belong to the European Union, Ireland joined in 1973 and Portugal in Firstly, a growth accounting analysis is carried out for both countries in order to give some insight about the behaviour of these economies and about the sources of their economic growth. In this analysis two estimation methods are applied: simple growth accounting and growth country regressions. Secondly, the importance of labour productivity as a growth determination factor is analysed. This is done performing a vector autoregressive analysis (VARs). More precisely, three VARs are estimated for each country, which will allow the study of the relationships between labour productivity and economic growth, labour productivity and living standards and finally labour productivity and inflation. Complementarily, an analysis of the correlation, the Impulse Responses and the Granger Causality between those pairs of variables (again for both countries) is performed. Finally, we compile the major lessons from Ireland and we leave some policy recommendations for Portugal. This paper is organised into five main sections. The next section is dedicated to the growth accounting for Portugal and Ireland. In section III, the importance of labour productivity for Portugal and Ireland is analysed through the VARs. Within each of these 2 sections a brief description of the data used is provided, the estimation method used is presented and the main results are 2

4 described and analysed. Section IV comprises the compilation of the main lessons from Irish performance and the suggestion of some policies which we believe to be fundamental for the Portuguese economy. Finally, in section V the main conclusions are presented. II. Growth Accounting In this section, a growth accounting analysis for Portugal and Ireland is performed, with the objective to acquire some insight about the economic growth of these countries. This analysis will be based on the neoclassical growth accounting. Before proceeding, some weaknesses of this theory should be mentioned. On the one hand, this theory does not explain the strengths that are behind capital and labour input (two growth sources). On the other hand, the so called total factor productivity (interpreted as technological progress) remains exogenous and unexplained by the model. In the next lines a short explanation is given about the neoclassical growth accounting. Output is assumed to grow through increases in the production factor inputs labour and capital and through improvements in technology. To investigate the sources of growth the following production function is used: Y = A F(K,L) (1) where Y denotes output, K denotes capital input, L represents labour input and A is total factor productivity (technological progress). Labour and capital are assumed to be homogeneous. The growth rate of the total factor productivity is given by the equation presented bellow: d ln Y = v K d ln K + v L d ln L + d ln A (2) where v K = (1 - v L ), is capital s share of national income and v L is labour s share of national income. The last term of equation (2) corresponds to the total factor productivity, which accounts for the growth not explained by capital accumulation or increased labour input (Solow s residual). The components 3

5 usually assumed to be included in this unexplained growth are: advances in knowledge, efficiency stand out, research, education and training. The capital stock can be estimated using the standard perpetual inventory method: K t = I t + (1. t-1 (3) Growth theory and empirical work on growth tell us that the growth of a country is mainly related with the following four aspects. First, the investment in physical capital, equipment and infrastructures is very important. Countries that invest a greater share and that have a high private investment tend to grow faster, even if only for a transitory period. Another very important issue is the investment in human capital, education and training. A third aspect is related to productivity growth. The literature suggests that free market policies (small governments with open markets that encourage foreign trade) are more likely to produce faster productivity growth. Finally, incentives (e.g. tax incentives) for research and development can also be determinant. Data As shown before, to perform the growth accounting the following data are needed: real output, labour input, investment, average depreciation rate and the capital s share of national income. Since the data to compute the real value of labour s or capital s share of income were not available, they were assumed to be approximately 70% and 30%, respectively (as we know labour s share of income is relatively larger than the capital s share). Computations were done with shares of 80% - 20% and 60% - 40% and the results did not change significantly. For the same reasons average depreciation rate was assumed to be 5%. The previous computations were done as well with a depreciation rate of 10% and again the results underwent insignificant modifications. 4

6 As a measure of labour input two series were used: employment and total hours of work. This last series was constructed from the number of weekly hours of work. Portuguese weekly hours of work were supplied by Portuguese Ministry of Social Security and Work, while Irish weekly hours of work were obtained through Datastream, OECD Economic Outlook. As a proxy for investment the Gross Fixed Capital Formation (GFCF) was used. Real GDP was used to measure real output. GFCF and GDP series were also collected from Datastream. Estimation Procedures Two methods were used to perform the growth accounting. The first one, the simple growth accounting analysis, consists in computing total factor productivity using the equations (2) and (3). The second method comprises the estimation of the following equation: d ln Y t = 1 2 d ln K t + 3 d ln L t t (4) where 2 denotes to the capital s share of national income, 3 applies for the ODERXU V VKDUH RI QDWLRQDO LQFRPH DQG t corresponds to the Total Factor Productivity (unexplained growth). Since the regression using first differences was unsatisfying for Portugal, the same regression was run in levels (using logged variables, without computing the first differences) plus a time trend. This procedure prevents the loss of information which might happen with the use of first differences. Results Until 1991 the behaviour of the Portuguese and Irish economies was similar. From 1991 the Irish GDP grew, in average, a 4,7% more per year than the Portuguese GDP (Chart 1 to 5). The results suggest that this increase was mainly due to an increase in the total factor productivity - accelerating total factor productivity growth has led to a notable growth revival (Chart 4 and 5). 5

7 In the 90 s the Irish growth rate of labour and capital inputs were also above the ones of Portugal (Chart 2 and 3), but their contribution to the increase in GDP was not as important as the contribution of total factor productivity (Chart 5). This is in accordance with the majority of the authors, whose works highlight technical progress. Of course capital investment is necessary since it is not possible to have technological improvement without new type of equipment. Indeed, Ireland had, in the last decade, a lot of investment from abroad which played an important role. Another important finding is that Ireland has faced an increased efficiency, which has been in the base of its economic growth in the last years. It is worth to note that until 1994 the percentage of the GDP growth not explained was always higher for Portugal than for Ireland (Chart 4). In 1994 this trend is reversed and in 2002 the percentage of the unexplained GDP growth was 2,5% for Portugal and 4,6% for Ireland. In what concerns the behaviour of the labour and capital inputs, between 1960 and 2002 employment has grown 52% in Portugal, whereas in Ireland it grew 67%. On the other hand, the capital stock in Portugal grew 462%, in the same period, while in Ireland it grew 384%. From the analysis of these numbers one can infer that the investment realised in Portugal was not as efficient as the one realised in Ireland (the investment does not seem to generate any increase in efficiency). Clearly, investment also plays a role for growth and development but it is not sufficient. By estimating equation 4 we expect to obtain positive coefficients for 2 and 3, moreover their values should be close to 0,3 and 0,7 respectively (the capital s share of national income and the labour s share of national income). Relatively to the estimation for Portugal 1 (Table 1), the first regression presents a low R 2, a negative sign for 3 (which was not expected) and despite that, this coefficient does not seems to be significant. On the other hand, 2 is significant at a 5% level but it is too high relatively to what was expected. Regressing employment and capital separately does not improve the results, moreover the 6

8 results are quite similar to the previous ones (in terms of goodness of fit, magnitude, significance level and signs of the coefficients). In fact, there is no evidence of a correlation between employment and capital (the correlation is only 7%). The regression in levels presents a very good R 2. 2 is statistically significant (at a 5% level) and it explains almost all the GDP growth. 3 is not statistically significant. For Ireland (Table 2), the value of 2 is lower than expected and it is not significant. The value of 3 is slightly higher than the prevision (but not so high as the one for Portugal) and it is very significant. The R 2 is not very high, but since we are working with first differences we can consider the result obtained very good. The Durbin-Watson statistic gives us evidence of no correlation in the residuals. These regressions confirm the results presented before: in the case of Ireland, the induced increase in efficiency is more important than the increase in capital growth and labour growth (Irish economic growth is mostly due to an increase in total factor productivity). In fact, the regression (Table 2) shows that 50% of the growth on output is not explained by growth in capital or in labour. In what concerns Portugal, again we verify that economic growth is explained, almost entirely, by the increase in the inputs. These extreme results for Portugal and Ireland confirm the different performance these two countries had over the 1990s. III. Importance of Labour Productivity Vector Autoregression Analysis This section is devoted to the importance of labour productivity growth. Before continuing it is worth to mention and describe the most common measures of productivity. One is average labour productivity, which is defined as real output per hour of work. In the present paper another definition for this 1 We have incorporated an AR(1) process in the error term to eliminate the correlation in the residuals (without the AR(1) term the Durbin-Watson statistic was low so we could not reject the hypothesis of serial correlation). 7

9 concept is also used: real output per employee 2. Finally, we still have total factor productivity. This concept is more complicated and it is defined as real output per unit of all inputs. As already mentioned before, it accounts for the growth not explained by capital accumulation or increased labour input. Total factor productivity is assumed to comprise advances in knowledge, efficiency stand out, technology, economies of scale, research, education and training. The concepts of labour productivity growth and total factor productivity by their own are not very meaningful. As Steindel and Stiroh (2001) argued, it is more interesting to analyse the relationship between those concepts and economic growth, growth in real per capita income, and inflation. To accomplish this objective we used Vector Autoregression analysis (VAR). Each VAR is performed with labour productivity growth and one of the following three variables: GDP growth, GDP per capita growth and inflation. Data To carry out the proposed vector auto regression analysis the following data is required, for Portugal and Ireland: output, population, inflation rate, employment and total number of hours worked. Employment and total number of hours worked are used as labour input measures. The minimum number of observations desirable to perform a vector autoregressive analysis is approximately eighty observations. For Portugal, it was possible to obtain quarterly data beginning in 1980, for the variables mentioned, except for population. Between 1980 and 1988 Portuguese population was only available in an annual basis. Hence, it was necessary to convert this annual data into quarterly data 3. On the other hand, for Ireland it was impossible to find quarterly data for the previous variables. According to the National Statistics Office they have started to collect quarterly data only after Therefore, it was necessary to change the frequency of all Irish 2 We have decided to work with productivity per employee because worked hours were not available in a quarterly basis for the desired period. 3 This was done using the spline procedure in E-Views 4.1 (multiplicative method). 8

10 series from annual (which were available) to quarterly 4. The only exception to this procedure was the series of weekly hours worked since it was available in a quarterly basis. For both countries this variable was accessible just from 1985 on. In what concerns the data that was converted from annual to quarterly their sources are the ones mentioned in section II-Data. The remaining variables were obtained from Datastream, however they have different sources: GDP and GDP deflator source is the International Financial Statistics and population and employment source is the National Statistics Office. As already mentioned before, Irish weekly hours worked source is the Central Statistics Office and the Portuguese weekly hours worked were obtained in the Portuguese Ministry of Social Security and Work. All the variables in growth rates are plotted in Chart 6, 7, 8, 9 and 10. For all the series, tests for the presence of seasonality were executed, and the series that showed evidence of seasonality were deseasonalised 5. The four tests available (through X12-EViews) are the followings: test for the presence of seasonality assuming stability, nonparametric test for the presence of seasonality assuming stability, moving seasonality test and the combined test for the presence of identifiable seasonality. A series was taken to be seasonal when two tests among the first three show evidence of seasonality. The series that have shown evidence of seasonality were: GDP and employment for Portugal and weekly hours of work for both countries. To obtain the growth rates of the variables logarithms were taken from the series and then first differences were taken from the logged data. In general, the resulting series are stationary. Nevertheless, to check, the Augmented Dickey-Fuller Test was executed for all the variables (Table 3 and 4). The results are that the Portuguese series are stationary at all significance levels, whereas for the Irish case just the log-difference of productivity per employee and per hour worked are stationary at all levels of significance. For the logdifference of Irish GDP and Irish GDP per capita the hypothesis of the existence of unit roots can be rejected only at a 10% significance level. Surprisingly, the 4 Again, this transformation was done using E-Views

11 Irish inflation rate is not stationary. Hence, it was necessary to do the second differences for this series (the resulting series is stationary at a 1% level). VAR Estimation As already mentioned, the VARs for both countries were performed with two different measures of productivity: GDP per hour worked and GDP per employee. In principle, total hours worked should be a more precise measure of productivity. However, the results obtained using this measure are worst than the results obtained using simply total employment (see Table 3). Besides that, for total hours worked the data is available only after Therefore, it was decided to continue the work only with productivity per employee. In order to choose the best lag structure specification the (unrestricted) VARs were performed with 2, 4, 6, 8 and 10 lags. Looking at the Akaike information criterion (which gives the best lag) and the Adjusted R 2 (which measures the validity of additional variables) (Table 5) we can see that in the case of Portugal the best results are obtained with 4 lags. Although, the results obtained with 6 lags are quite similar. In what concerns Ireland, for the VAR with productivity growth and GDP growth the best result is obtained with 10 lags and for the other regressions the best results are obtained with 6 lags. Since for Portugal the results do not differ a lot with 4 or 6 lags and for Ireland 6 lags seems to be the best choice, for the sake of simplicity all the further analysis were based in the VARs with 6 lags. From the economic point of view this also seemed an adequate choice since 6 lags correspond to 1 year and a half. Consider a larger period does not appeared to be necessary because it is not likely that the variables will exert some effect after 1 year and a half and consider a smaller period would be too restrictive. After analysing the relationships between productivity growth and the other three variables, the study proceeds with the analysis of: the Impulse 5 Using the X12, procedure E-Views. 10

12 Responses 6, the Granger Causality Test 7, and of the Variance Decomposition 8, for the chosen models (productivity per employee and VARs with 6 lags). Results First of all, it is worth to examine the evolution of labour productivity between 1980 and In Chart 9, we can observe that Portuguese and Irish growth of the labour productivity per employee have a regular evolution over the years. The only aspects to stress are that the Portuguese productivity growth has a negative peak in 1983 and two positive peaks in 1988 and However, these peaks just stand by one year and are not observed thereafter. Chart 10, shows a very similar evolution for productivity per hour worked, in the case of Portugal. In the Irish case, the growth of the productivity per hour worked shows some peaks that before were not observed. One is negative and it occurs in 1988 and the other one is positive and it occur in But again it is not possible to distinguish medium/long periods of productivity slowdown or of a strengthening in productivity growth. (i) Productivity Growth and Economic Growth Without any previous analysis one would say that productivity growth and economic growth are closely related. By definition, output growth is the sum of the growth of labour hours and of labour productivity growth (Steindel and Stiroh, 2001). Thus, a higher labour productivity growth seems to imply a higher GDP growth. When a long period is analysed demographic changes can affect output growth independently of productivity growth. Observing Chart 11, which presents the Portuguese growth rate of output and the Portuguese labour productivity for each decade and each half decade, we can verify that GDP growth and productivity growth follow the 6 The Impulse Response function shows the effect that a shock in a specific moment and variable has on current and future values of the endogenous variables. 7 The Granger Causality Test tests whether an endogenous variable can be treated as exogenous. 8 The Variance Decomposition accounts for the proportion of the variance of an endogenous variable due to the different structural shocks. 11

13 same path; no matter whether we look at the first or second chart. As expected, decreases/increases in productivity growth are accompanied by decreases/increases in economic growth. Nevertheless, when we look at Chart 12, which plots the same information but now for Ireland, there is no evidence of a relationship between those two variables. The most illustrative example of this lack of link are the 90s, where output growth doubles relatively to the output growth in the 80s and the productivity growth decreases slightly. In the second half of the 70 s again productivity decreases whereas the output growth increases. To get some insight about the effects of time in the relation between productivity growth and output growth the correlation between these two variables is plotted in Chart 13. For Portugal it is possible to observe a high correlation in the first quarter that vanishes quite fast. In the long run no significant correlation is detected. In what concerns Ireland, a short run correlation is also observed, however it is not as strong as the one before. Surprisingly, between the fifteenth quarter and the eighteenth quarter we observe a considerable negative correlation between Irish productivity growth and output growth. It is worth to refer that for the United States Steindel and Stiroh (2001) have also found this short term relationship, followed by a flat correlation up to five years, which then rises. (ii) Productivity Growth and Per-Capita Income Growth The main justification to believe in a relationship between productivity growth and income growth relies on the statement, defended by many authors, that the productivity growth and the real wages are equal (Steindel and Stiroh, 2001). In fact, when we observe Chart 14, either for Portugal or Ireland, this relationship does not seem to exist. For each country the evolution of the correlation between per-capita income growth and productivity growth is very close to the evolution of the correlation between output growth and productivity growth. This result goes completely against to what is found by Steindel and Stiroh (2001) for the United States. Indeed, they found a very weak correlation 12

14 in the short term (which for Portugal and Ireland is strong) and they justify this fact with no constant returns to scale, with the slippage between growth in wages per hour and growth in income per-capita and with the inexistence of fixed relationship between hours worked and population. They argue that in the long term these divergences tend to disappear, which explains the increasing correlation between per-capita income and productivity growth that is observed for the United States. However, for Portugal and Ireland this is not the case. (iii) Productivity Growth and Inflation In principle, there is no obvious reason for the existence of a relationship between these two variables. However, we can see, in the Portuguese data (Chart 15), a negative correlation in the short run between productivity growth and inflation. Thereafter, it does not exists a significant correlation between these two variables. In the case of Ireland (Chart 15), in the short run no important correlation is observed, but in the third year and a half it starts to arise a positive correlation between productivity growth and inflation, which decreases through time and reaches the value of -5,5% in the sixth year. For the United States, Steindel and Stiroh (2001) found persisting negative correlation between productivity growth and inflation ( it seems to be the case that periods of higher productivity growth are periods of lower inflation 9 ). The justifications presented for this fact are: a) since higher inflation rates can distort the price mechanism they can also reduce the economic efficiency, having a negative impact on capital accumulation and technological progress; b) since the periods of high productivity growth are periods of relatively fast economic growth, the monetary authorities can profit and implement anti-inflationary policies. 9 Steindel and Stiroh (2001). 13

15 (iv) VAR Estimations and Impulse Responses All the VAR estimations for Portugal are very poor. The R 2 and Adjusted R 2 are very low, the coefficients are not statically significant and the F-statistic is also very low in all the regressions. The reason for this may be just the lack of variability in the data. For Ireland the VAR estimation results are always very good. The R 2 and Adjusted R 2 are high, the F-statistic is also high and some of the coefficients show a high level of significance. The impulse responses allow us to clarify the linkages between productivity growth and each of the other variables. For Portugal (Chart 16), a shock in productivity does not influence significantly either the output growth, or the per-capita income growth. On the other hand a shock on GDP growth, or in per-capita income growth, has an instantaneous influence on productivity, but this relation quickly disappears. For Ireland (Chart 17), the short-term impulse response relations from productivity growth to output growth and per-capita income growth and from output growth and per-capita income growth to productivity growth are weak in magnitude. For Ireland we observe smooth impulse responses, which are not very common. This is related with the fact that the quarterly data for Ireland is obtained from annual data. To check if this fact was having any influence in the magnitude of the impulse responses this analysis was performed also with productivity per hour worked ( truly quarterly data). With this productivity measure the magnitude of the shocks does not change significantly. The change is moreover in the pattern of the impulse responses, which comes closer to what we observe for Portugal. The influence of productivity on inflation, in the Irish case, is very weak. Nevertheless, inflation seems to influence productivity two years after the shock; however this is not a very strong relation. This is also the case for Portugal, with the only difference that the influence of inflation on productivity appears 16 months after the shock on inflation. (v) Granger Causality Test 14

16 Using the Granger Causality test (Table 12) it is possible to see that for Portugal there is no evidence of influence of productivity growth on economic growth or vice-versa. In the case of Ireland (Table 13) productivity growth predicts GDP growth at the 1% level and GDP growth predicts productivity growth but just at the 5% level. These results are consistent with what we have found with growth accounting analysis, i.e., once more we have evidence that Irish economic growth is largely due to increasing productivity whereas for Portugal this is not observable. The results of the Granger Causality test for productivity growth and per-capita income growth are very similar to the ones presented for productivity growth and economic growth. Once more, for Portugal, there is no evidence of influence of productivity growth on per capita income growth, neither from this one on productivity growth. For Ireland, the causality flow is again bidirectional. However, the magnitude of the flow from productivity growth to per-capita income growth is stronger than in the reverse direction. When productivity growth and inflation are considered the result from the Granger Causality test is similar for both countries: there is no evidence of prediction between productivity growth and inflation or vice-versa. IV. Lessons for Portugal In this section we summarize the main lessons we can withdraw from the evolution of the Irish economy and we present some policy recommendations for Portugal. The strong Irish economic growth in the 1990s was mainly due to the interaction between the following factors: increase in labour supply and in labour productivity, increase in foreign investment and as well a long period of social consensus and a favourable fiscal regime. The importance of two former factors is not observable from the analysis presented in section II and III nevertheless they also contributed for the Irish economic growth. The increase in the quantity and quality of labour supply was determinant. This increase results from the entrance in the active population of a 15

17 large number of young people with relatively high qualifications. Nowadays, Ireland is among the OECD countries with a larger number of scientists and engineers since almost half of the students go to the university. As we have mentioned before the foreign investment also played an important role in the economic revival. The most modern sector in the economy (computing, chemistry and drinks without alcool) grew more than 2 thirds between 1994 and Moreover, only the foreign companies working in the sector of manufactory produce approximately 30% for GDP. The social consensus among social partners (namely the Program of Competitiveness and Work signed in 1996) was also important since it made possible to sustain the Irish competitiveness relatively to her commercial partners. Finally, the moderation in wages achieved through healthy budget and monetary policies and efficient revenue management was as well important for the Irish economic growth. We already stated that one of the biggest problems that Portugal faces is the lack of qualifications of the labour market. To illustrate the magnitude of the problem we report the percentage of the population between 25 and 64 years old, for Portugal and for Ireland, that has attained at least upper secondary education in In Portugal only 18% of males and 22% of females have attained the referred degree of education while in Ireland the numbers are respectively, 48% and 54%. Despite the scenario for Ireland being much better than for Portugal still Ireland remains in the second half of the OECD ranking, (which comprises 28 countries). It is urgent that Portugal approaches his level of education and professional formation to the average level of OECD countries. Some measures concerning labour productivity that we think to be important to implement in Portugal are the followings: x reduce the number of students that give up from school prematurely; x extend the possibilities of formation for those who do not intend to continue one s studies; x create formation programs addressed to the active population (this competence should be shared with the employers); 16

18 x x x create and promote technical education considering the country needs and direct the university degrees to the prior areas in the country; create more mechanisms to pass from the school system to the active life; invest in technology and research. V. Conclusions We have applied three methods of estimation to investigate the differences in the Portuguese and Irish economic performance. Those methods were: simple growth accounting, growth country regressions and VARs by country. These three procedures lead us to same conclusion: Irish economic growth is largely due to increasing (in the 1990s) total factor productivity (TFP) whereas in Portugal changes in labour and capital account almost entirely for the evolution of real GDP. When we analyse more closely the characteristics of the labour and investment in both countries we realize that Ireland has had the great capacity of increasing her efficiency through advances in education, training, knowledge, technology, research and use of labour intensively. This has been the principal engine of the Irish growth and it seems to be what is missing to Portugal. It is very urgent that Portugal approaches its level of education and professional formation to the average level of OECD countries in order to achieve a higher labour productivity. The success of the Irish strategy to attract foreign investment has also stimulated the economy in a very important way. Clearly, investment also plays a role for growth and development but it is not sufficient. It must be carried through the right sectors to generate the desired effects. The type of investment made in Portugal in the last decade does not seem to be very fruitful in the long run. The correlation analysis between the two European countries under study, on the one hand, and the United States in the background article by 17

19 Steindel and Stiroh (2001), on the other hand, does not allow us to take conclusions and would need a further study. 18

20 REFERENCES Bourbonnais, Régis, 2000, Économétrie Manuel et exercices corrigés, 3 rd edition, Dunod Cornwall, John and Cornwall, Wendy, 2001, A demand and supply analysis of productivity growth, Journal of Economic Literature Études Économiques de L OCDE, 1976, Portugal Études Économiques de L OCDE, , Portugal Études Économiques de L OCDE, 1998, Portugal Études Économiques de L OCDE, 1999, Portugal Études Économiques de L OCDE, 1972, Ireland Études Économiques de L OCDE, 1973, Ireland Études Économiques de L OCDE, 1974, Ireland Études Économiques de L OCDE, , Ireland Études Économiques de L OCDE, 1999, Ireland J. Stiroh, 2001, What drives productivity growth?, FRBNY Economic Policy Review Jorgenson, Dale W. and Stiroh, Kevin J., 2000, Raising the Speed Limit: U.S. Economic Growth in the Information Age, Journal of Economic Literature Jorgenson, Dale W., 1995, Productivity International Comparisons of Economic Growth, MIT Press Lambelet, Jean-Christian and Mihailov, Alexander, 1999, A note on Switzerland s economy: did the Swiss economy really stagnate in the 1990s, and is Switzerland all that rich?, Analyses & Prévisions, Institute Créa of Applied Macroeconomics, University of Lausanne 19

21 Romer, David, 2001, Advanced Macroeconomics, 2 nd edition, McGraw-Hill Steindel, Charles and Stiroh, Kevin J., 2001, Productivity: What is it, and why do we care about it?, Journal of Economic Literature 20

22 APPENDIX 1 Chart 1: GDP growth for Portugal (on the left) and Ireland (on the right) - year-over-year percentage change DLNGDP_IR Chart 2: Employment growth for Portugal (on the left) and Ireland (on the right) - year-over-year percentage change DLNEMP_PT Chart 3: Capital growth for Portugal (on the left) and Ireland (on the right) - year-over-year percentage change DLNCAPITAL_PT 21

23 Chart 4: Total factor productivity for Portugal and Ireland (using employment to measure labour contribution to growth) Total factor productivity Years PT IR Chart 5: GDP index decomposition for Portugal and Ireland (using employment to measure labour contribution to growth) 8 Portugal index (1960=1) GDP index labor contribution to GDP labor + capital contribution to GDP Ireland index (1960=1) GDP index labor contribution to GDP labor + capital contribution to GDP 22

24 Table 1: Growth Accounting Estimations for Portugal ( ) Dependent Variable: DLNGDP_PT Method: Least Squares Date: 07/29/03 Time: 14:19 Sample(adjusted): Included observations: 42 after adjusting endpoints Convergence achieved after 19 iterations Variable Coefficient Std. Error t-statistic Prob. C DLNCAPITAL_PT DLNEMP_PT AR(1) R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic) Inverted AR Roots.27 Dependent Variable: DLNGDP_PT Method: Least Squares Date: 07/29/03 Time: 14:21 Sample(adjusted): Included observations: 42 after adjusting endpoints Convergence achieved after 16 iterations Variable Coefficient Std. Error t-statistic Prob. C DLNCAPITAL_PT AR(1) R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic) Inverted AR Roots.30 23

25 Dependent Variable: DLNGDP_PT Method: Least Squares Date: 07/29/03 Time: 14:22 Sample(adjusted): Included observations: 42 after adjusting endpoints Convergence achieved after 11 iterations Variable Coefficient Std. Error t-statistic Prob. C DLNEMP_PT AR(1) R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic) Inverted AR Roots.38 Dependent Variable: LNGDP_PT Method: Least Squares Date: 07/29/03 Time: 14:23 Sample(adjusted): Included observations: 43 after adjusting endpoints Convergence achieved after 37 iterations Variable Coefficient Std. Error t-statistic Prob. C LNCAPITAL_PT LNEMP_PT AR(1) R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic) Inverted AR Roots.88 24

26 Table 2: Growth Accounting Estimations for Ireland ( ) Dependent Variable: DLNGDP_IR Method: Least Squares Date: 07/25/03 Time: 13:13 Sample(adjusted): Included observations: 43 after adjusting endpoints Variable Coefficient Std. Error t-statistic Prob. C DLNCAPITAL_IR DLNEMP_IR R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic)

27 APPENDIX 2 Chart 6: GDP growth for Portugal (on the left) and Ireland (on the right) quarter over quarter percentage change DLNGDP_IR Chart 7: GDP per capita growth for Portugal (on the left) and Ireland (on the right) quarter over quarter percentage change DLNGDPPERCAPITA_IR Chart 8: Inflation for Portugal (on the left) and Ireland (on the right) quarter over quarter percentage change INFLATION_IR 26

28 DINFLATION_IR Chart 9: Productivity (per employee) growth for Portugal (on the left) and Ireland (on the right) quarter over quarter percentage change DLNPRODUCTIVITYE_IR Chart 10: Productivity (per hour worked) growth for Portugal (on the left) and Ireland (on the right) quarter over quarter percentage change DLNPRODUCTIVITYH_PT DLNPRODUCTIVITYH_IR 27

29 Table 3: Augmented Dickey-Fuller Unit Root Test for the Portuguese Series Null Hypothesis: DLNGDP_PT_SA has a unit root Exogenous: Constant Lag Length: 0 (Automatic based on SIC, MAXLAG=11) t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Null Hypothesis: DLNGDPPERCAPITA_PT has a unit root Exogenous: Constant Lag Length: 0 (Automatic based on SIC, MAXLAG=11) t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Augmented Dickey-Fuller Test Equation Dependent Variable: D(DLNGDP_PT_SA) Method: Least Squares Date: 05/03/03 Time: 20:42 Sample(adjusted): 1980:3 2002:2 Included observations: 88 after adjusting endpoints Variable Coefficient Std. Error t-statistic Prob. DLNGDP_PT_SA(-1) C R-squared Mean dependent var 2.99E-05 Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic) Augmented Dickey-Fuller Test Equation Dependent Variable: D(DLNGDPPERCAPITA_PT) Method: Least Squares Date: 05/03/03 Time: 20:41 Sample(adjusted): 1980:3 2002:2 Included observations: 88 after adjusting endpoints Variable Coefficient Std. Error t-statistic Prob. DLNGDPPERCAPITA_PT(-1) C R-squared Mean dependent var 3.85E-05 Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic) Null Hypothesis: DINFLATION_PT has a unit root Exogenous: Constant Lag Length: 0 (Automatic based on SIC, MAXLAG=11) t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Null Hypothesis: DLNPRODUCTIVITYE_PT has a unit root Exogenous: Constant Lag Length: 0 (Automatic based on SIC, MAXLAG=11) t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Augmented Dickey-Fuller Test Equation Dependent Variable: D(DINFLATION_PT) Method: Least Squares Date: 05/03/03 Time: 20:42 Sample(adjusted): 1980:3 2002:2 Included observations: 88 after adjusting endpoints Variable Coefficient Std. Error t-statistic Prob. DINFLATION_PT(-1) C R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic) Augmented Dickey-Fuller Test Equation Dependent Variable: D(DLNPRODUCTIVITYE_PT) Method: Least Squares Date: 05/03/03 Time: 20:41 Sample(adjusted): 1980:3 2002:2 Included observations: 88 after adjusting endpoints Variable Coefficient Std. Error t-statistic Prob. DLNPRODUCTIVITYE_PT(-1) C R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic) Null Hypothesis: DLNPRODUCTIVITYH_PT has a unit root Exogenous: Constant Lag Length: 0 (Automatic based on SIC, MAXLAG=10) t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Augmented Dickey-Fuller Test Equation Dependent Variable: D(DLNPRODUCTIVITYH_PT) Method: Least Squares Date: 05/03/03 Time: 20:36 Sample(adjusted): 1985:3 2002:2 Included observations: 68 after adjusting endpoints Variable Coefficient Std. Error t-statistic Prob. DLNPRODUCTIVITYH_PT(-1) C R-squared Mean dependent var -2.35E-05 Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic)

30 Table 4: Augmented Dickey-Fuller Unit Root Test for the Irish Series Null Hypothesis: DLNGDP_IR has a unit root Exogenous: Constant Lag Length: 9 (Automatic based on SIC, MAXLAG=11) t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Null Hypothesis: DLNGDPPERCAPITA_IR has a unit root Exogenous: Constant Lag Length: 9 (Automatic based on SIC, MAXLAG=11) t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Augmented Dickey-Fuller Test Equation Dependent Variable: D(DLNGDP_IR) Method: Least Squares Date: 05/03/03 Time: 20:42 Sample(adjusted): 1982:4 2002:4 Included observations: 81 after adjusting endpoints Variable Coefficient Std. Error t-statistic Prob. DLNGDP_IR(-1) D(DLNGDP_IR(-1)) D(DLNGDP_IR(-2)) D(DLNGDP_IR(-3)) D(DLNGDP_IR(-4)) D(DLNGDP_IR(-5)) D(DLNGDP_IR(-6)) D(DLNGDP_IR(-7)) D(DLNGDP_IR(-8)) D(DLNGDP_IR(-9)) C R-squared Mean dependent var 3.72E-05 Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid 5.94E-05 Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic) Augmented Dickey-Fuller Test Equation Dependent Variable: D(DLNGDPPERCAPITA_IR) Method: Least Squares Date: 05/03/03 Time: 20:42 Sample(adjusted): 1982:4 2001:4 Included observations: 77 after adjusting endpoints Variable Coefficient Std. Error t-statistic Prob. DLNGDPPERCAPITA_IR(-1) D(DLNGDPPERCAPITA_IR(-1)) D(DLNGDPPERCAPITA_IR(-2)) D(DLNGDPPERCAPITA_IR(-3)) D(DLNGDPPERCAPITA_IR(-4)) D(DLNGDPPERCAPITA_IR(-5)) D(DLNGDPPERCAPITA_IR(-6)) D(DLNGDPPERCAPITA_IR(-7)) D(DLNGDPPERCAPITA_IR(-8)) D(DLNGDPPERCAPITA_IR(-9)) C R-squared Mean dependent var 5.73E-05 Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid 6.37E-05 Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic) Null Hypothesis: DINFLATION_IR has a unit root Exogenous: Constant Lag Length: 4 (Automatic based on SIC, MAXLAG=11) t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Augmented Dickey-Fuller Test Equation Dependent Variable: D(DINFLATION_IR) Method: Least Squares Date: 05/03/03 Time: 20:43 Sample(adjusted): 1981:3 2002:4 Included observations: 86 after adjusting endpoints Variable Coefficient Std. Error t-statistic Prob. DINFLATION_IR(-1) D(DINFLATION_IR(-1)) D(DINFLATION_IR(-2)) D(DINFLATION_IR(-3)) D(DINFLATION_IR(-4)) C R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic) Null Hypothesis: DDINFLATION_IR has a unit root Exogenous: Constant Lag Length: 3 (Automatic based on SIC, MAXLAG=11) t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Augmented Dickey-Fuller Test Equation Dependent Variable: D(DDINFLATION_IR) Method: Least Squares Date: 05/13/03 Time: 18:06 Sample(adjusted): 1981:3 2002:4 Included observations: 86 after adjusting endpoints Variable Coefficient Std. Error t-statistic Prob. DDINFLATION_IR(-1) D(DDINFLATION_IR(-1)) D(DDINFLATION_IR(-2)) D(DDINFLATION_IR(-3)) C R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic) Null Hypothesis: DLNPRODUCTIVITYE_IR has a unit root Exogenous: Constant Lag Length: 9 (Automatic based on SIC, MAXLAG=11) t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Null Hypothesis: DLNPRODUCTIVITYH_IR has a unit root Exogenous: Constant Lag Length: 0 (Automatic based on SIC, MAXLAG=10) t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Augmented Dickey-Fuller Test Equation Dependent Variable: D(DLNPRODUCTIVITYE_IR) Method: Least Squares Date: 05/03/03 Time: 20:41 Sample(adjusted): 1982:4 2002:4 Included observations: 81 after adjusting endpoints Variable Coefficient Std. Error t-statistic Prob. DLNPRODUCTIVITYE_IR(-1) D(DLNPRODUCTIVITYE_IR(-1)) D(DLNPRODUCTIVITYE_IR(-2)) D(DLNPRODUCTIVITYE_IR(-3)) D(DLNPRODUCTIVITYE_IR(-4)) D(DLNPRODUCTIVITYE_IR(-5)) D(DLNPRODUCTIVITYE_IR(-6)) D(DLNPRODUCTIVITYE_IR(-7)) D(DLNPRODUCTIVITYE_IR(-8)) D(DLNPRODUCTIVITYE_IR(-9)) C R-squared Mean dependent var 2.57E-05 Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid 5.41E-05 Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic) Augmented Dickey-Fuller Test Equation Dependent Variable: D(DLNPRODUCTIVITYH_IR) Method: Least Squares Date: 05/03/03 Time: 20:41 Sample(adjusted): 1985:3 2000:4 Included observations: 62 after adjusting endpoints Variable Coefficient Std. Error t-statistic Prob. DLNPRODUCTIVITYH_IR(-1) C R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood F-statistic Durbin-Watson stat Prob(F-statistic)