We study how labour migration modified the Dutch disease effects during the Norwegian resource boom 2004–2013. In these years the resource movement effect of the petroleum industry was larger than the spending effect. This was mainly due to the introduction of a fiscal policy rule in 2001 that limited spending. The EU-enlargement in 2004 increased labour migration and affected also the Norwegian labour market. We find that economic growth in Norway was roughly doubled from 2004 to 2013 because of the resource boom while total population increased by 2% because of higher immigration. Moreover, both the resource movement and spending effects on Mainland GDP were roughly unaffected by immigration while employment increased, real wages fell and so did productivity. The negative effects of the boom on industries producing tradables were counteracted by endogenous terms of trade effects, immigration and demand effects of the boom.
This is a preview of subscription content, access via your institution.
Buy single article
Instant access to the full article PDF.
Tax calculation will be finalised during checkout.
Subscribe to journal
Immediate online access to all issues from 2019. Subscription will auto renew annually.
Tax calculation will be finalised during checkout.
New EU consists of 11 EU countries that joined the EU from May 2004 and onwards. Old EU is the rest of the EU countries plus EFTA countries, Australia, New Zealand, US and Canada. These groups coincide with country group 1 and 2 in the Norwegian population forecast, available at ssb.no. Other countries in the world are categorized in country group 3 and called ROW in Figure 4.
Most importantly, government policy and world market features are not affected by the changes in our counterfactual simulations.
A close inspection of the numbers in Figure 5 reveals that accumulated immigration is much larger than the change in population. This is mainly because some immigrants emigrate after a period of residence. Emigration rates by country group, age and gender are exogenous in the model making total emigration endogenous.
Results including this dummy are available upon request.
AR is a test for up to 2nd order residual autocorrelation used by Harvey (1981), ARCH is the Engle (1982) test for first order autoregressive conditional heteroskedasticity, NORM is the normality test described in Doornik and Hansen (2008), HET and HET-X are tests for residual heteroskedasticity due to White (1980) and RESET is a test for functional form misspecification based on Ramsey (1969).
Anundsen AK, Jansen ES (2013) Self-reinforcing effects between housing prices and credit. J Hous Econ 22:192–212
Banerjee A, Dolado JJ, Mestre R (1998) Error-correction mechanism tests for cointegration in a single-equation framework. J Time Ser Anal 19:267–283
Beine M, Bos CS, Coulombe S (2012) Does the Canadian economy suffer from Dutch disease? Resour Energy Econ 34:468–492
Beine M, Coulombe S, Vermeulen WN (2015) Dutch disease and the mitigation effect of migration: evidence from Canadian provinces. Econ J 125:1574–1615
Boug P, Fagereng A (2010) Exchange rate volatility and export performance: a cointegrated VAR approach. Appl Econ 42:851–864
Boug P, Cappelen Å, Eika T (2013) Exchange rate pass-through in a small open economy: the importance of the distribution sector. Open Econ Rev 24(5):853–879
Bowitz E, Fæhn T, Grunfeld LA, Moum K (1997) Can a wealthy economy Gain from EU-membership? Adjustment costs and long-term effects of full integration. The Norwegian case. Open Econ Rev 8(3):211–231
Bratsberg B, Raaum O (2012) Immigration and wages: evidence from construction. Econ J 122:1177–1205
Cappelen Å, Skjerpen T (2014) The effect on immigration of changes in regulations and policies: a case study. J Common Mark Stud 52:810–825
Cappelen Å, Skjerpen T, Tønnessen M (2015) Forecasting immigration in official population projections using an econometric model. Int Migr Rev 49:945–980
Chirinko RS (2008) σ: the Lond and short of it. J Macroecon 30(2):671–686
Corden WM (1984) Booming sector and Dutch disease economics: survey and consolidation. Oxf Econ Pap 36:359–380
Doornik JA, Hansen H (2008) An omnibus test for univariate and multivariate normality. Oxf Bull Econ Stat 70, Supplement:927–939
Engle RF (1982) Autoregressive conditional heteroscedasticity with estimates of the variance of United Kingdom inflation. Econometrica 50:987–1007
Fedotenkov I, van Groezen B, Meijdam L (2014) Demographic change, international trade and capital flows. Open Econ Rev 25:865–883
Gjelsvik, M. L., R. Nymoen, V. Sparrman (2015), Have inflation targeting and EU labour immigration changed the system of wage formation in Norway?., Disc Papers No824, Statstics Norway
Harbo I, Johansen S, Nielsen B, Rahbek A (1998) Asymptotic inference of cointegration rank in partial systems. J Bus Econ Stat 16:388–399
Harvey AC (1981) The econometric analysis of time series. Philip Allan, Oxford
Hungnes H (2011) A demand system for inputs factor when there are technological changes in production. Empir Econ 40:581–600
Jansen ES (2013) Wealth effects on consumption in financial crisis: the case of Norway. Empir Econ 45:873–904
Johansen S (1995) Likelihood-based inference in Cointegrated vector autoregressive models. Oxford University Press, New York
Layard R, Nickell S, Jackman R (2005) Unemployment. Macroeconomic Performance and the Labour Market, second edition. Oxford University Press, Oxford
Maddock R, McLean I (1984) Supply-side shocks: the case of Australian gold. J Econ Hist 44(4):1047–1067
Mayda, A.M. (2010), “International migration: a panel data analysis of the determinants of bilateral flows”, J Popul Econ, 23 (4), 1249–74, 1274
Pesaran MH, Shin Y, Smith RJ (2001) Bound testing approaches to the analysis of level relationships. J Appl Econ 16:289–326
Ramsey JP (1969) Tests for specification errors in classical linear least-squares regression analysis. J R Stat Soc Ser B 31:350–371
von Brasch T, Gjelsvik ML, Sparrman V (2018) Deunionization and job polarization – a macroeconomic model analysis for a small open economy. Econ Syst Res 30(3):380–399
White H (1980) A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 48:817–838
The project is financed by the Research Council of Norway, PETROSAM 2, grant 233687 (PROSPECTS). The authors thank T. Skjerpen, P. Boug and T. von Brasch for extensive comments on an earlier draft and two anonymous referees for their useful comments. All remaining errors are ours.
Conflict of Interest
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix: Modelling immigration
Appendix: Modelling immigration
Our starting point for modelling immigration to Norway is
Here Ii is immigration from region or country group i (i = 1,2,3) to Norway divided by total population in region 1. The Y’s are GDP per capita in PPPs in Norway and region i and the U’s are unemployment rates in Norway and region i. Dpol is a set of step-dummies that are included in order to capture possible policy regime changes related to immigration. For region 1 “Western countries” the only relevant dummy is a step dummy from 1994 and onwards when Norway joined the European Economic Area. This dummy was, however, not significant in any of the estimated specifications.Footnote 4 Data on population are taken from the UN while income data and unemployment rates are readily available from the OECD web page, cf. Cappelen et al. (2015).
Because multiple long-run relationships may exist among the variables included in the model of immigration, we employ the Johansen (1995, p. 167) trace test for cointegration rank determination. We thus start with an unrestricted p-dimensional VAR of order k having the form
where Yt is a (p × 1) vector of modelled variables (I, YN/Yi, UN, Ui) at time t, μ is a vector of intercepts, Π1, ...,Πk are (p x p) coefficient matrices of lagged level variables, t captures a trend and εk + 1, ..., εT are independent Gaussian errors with expectation zero and (unrestricted) (p x p) covariance matrix Ω. The initial observations Y1, ..., Yk are kept fixed.
The question now is how eq. (3) can be reparameterised to a cointegrated VAR (henceforth CVAR) in which immigration behaviour can be formulated as a reduced rank restriction on the impact matrix Π = −(I − Π1 − ... − Πk). The way the CVAR is formulated in our context depends on the exogeneity status of the unemployment series.
If Yt is I(1), then the first difference ΔYt is I(0), implying either Π = 0 or Π has reduced rank such that Π = αβ’, where α and β are (p x r) matrices and 0 < r < 4. Here r denotes the rank order of Π. Assuming for notational simplicity that k = 2, the CVAR becomes
where β’lnYt-1 is a (r × 1) vector of stationary cointegration relations among ln(I1)t, ln(YN /Y1)t, ln(UN)t and ln(U1)t and Γ1 = −Π2 is the (4 X 4) coefficient matrix of the lagged differenced variables.
We present the case where the two unemployment rates (the U’s in (1)) enter in relative form similar to relative incomes and are weakly exogenous for the long run parameters. In this case, valid inference on β can be obtained by considering the two-dimensional system of ln(I1)t and ln(YN/Y1)t conditional on log of relative unemployment (U) without loss of information, see Johansen (1995). Following Harbo et al. (1998), we may formulate the partial CVAR equivalent to (4) as (again assuming k = 2)
where ω and γ1 are (2 × 1) and (2 × 2) matrices, respectively. The corresponding marginal model is ΔlnUt = γ2 ΔlnYt-1 + δ2t + μ2 + ε2,t where γ2 is a (1 × 2) vector. It follows that the log of relative Ut is included in the long-run part of (5) as a non-modelled variable. As noted by Harbo et al. (1998), the asymptotic distribution of the trace test statistic is influenced by conditioning on weakly exogenous variables and standard critical values are thus not valid. We therefore use the critical values in Table 2 in Harbo et al. (1998). The constants enter the partial system unrestrictedly.
We find that using k = 2 produces a model which passes all single-equation and vector diagnostics according to standard tests. Table 2 reports trace test statistics for the partial CVAR.
The null hypothesis of no cointegration cannot be rejected according to the test in Harbo et al. (1998) at 5%. However, we show below that single equation tests for cointegration based on Banerjee et al. (1998) as well as Pesaran et al. (2001) lend some support for a cointegrating relationship between the immigration rate, relative incomes and relative unemployment.
Assuming the rank to be one, we may specify one restricted cointegrating vector normalized on the immigration rate (I1) as depending on relative incomes (YN/Y1) and relative unemployment rates as shown in (6) below. Given that the rank equals one, we test if relative income is weakly exogenous for the cointegrating relationship and that the trend is not significant. With χ2 (2) = 0.41 (p value = 0.52) we find that this joint hypothesis cannot be rejected. Thus, we proceed to single equation estimation based on the following cointegrating relationship
The estimate of the corresponding adjustment coefficient is −0.436 with an estimated standard error of 0.115 so the corresponding t-value equals −3.79. We thus define an equilibrium correction term (eqcm)
Using a general-to-specific modelling strategy we find the following well-specified parsimonious model using a sample from 1972 to 2016
The specification tests show no sign of any residual autocorrelation or heteroscedasticity and the error term is well proxied by Gaussian errors.Footnote 5 Note that the model includes no dummies at all. The parameter estimates are highly significant. The highly significant parameter estimate for the eqcm-term (t-value of - 4.2) supports cointegration. According to Table 1 in Banerjee et al. (1998) the critical value is – 3.28 supporting the existence of a cointegrating relationship. A generalization of this approach within a single equation framework is found in Pesaran et al. (2001) for both I(0) and I(1) regressors. This method consists of estimating an error correction form of an autoregressive dirstributed lag model and testing whether variables in levels could be deleted from the model or not. The maintained model is one in levels and we test if cointegration could be rejected or not. If H0 is not rejected there is support for cointegration. In our case we do not impose the results in eq. (7) but let lagged levels of the immigration rate, relative incomes and unemployment enter a model in first differences and lagged differences in line with our cointegration analyses earlier where we found that a VAR in levels with two lags provided a parsimonious representation with desirable statistical properties and in particular no autocorrelation. The resulting F-test has as a non-standard distribution and in our case the F-value from comparing a model with and without regressors in levels is 4.63. At 5% and in the case with no trend and unrestricted intercept the upper limit is 4.85 (cf. Table CI(iii) in Pesaran et al. 2001) for I(1)-variables). At 5% our test is in the inconclusive regions while if we look at the 10% level the F-statistics is 4.14 supporting cointegration. A test of whether or not the coefficient of the lagged dependent variable is zero is −3.45 while the critical value at 5% (cf. Table CII(iii)) is – 3.53. At 10% the critical value is 3.21 again lending support for cointegration. Recursive estimates (available upon request) show that eq. (8) has very stable parameters from 2003 and onwards and Chow tests do not indicate any breaks. Equation (8) is included in the multisector macromodel used in the simulations in this study.
Immigration from Eastern Europe (new EU-members in 2004 and 2007) is modelled in the same way as in (8). In this case we have not carried out a multivariate cointegration analysis because of too few observations. The breakdown of the Soviet-block in 1989 changed both economic and political conditions in these countries and comparable data from 1970 are not available. Thus, we use a sample from 1990 to 2015. Due to changes in immigration rules affecting countries in this region we include step dummies from 2004 and 2007 to capture the new policy regimes for immigration.
The model for immigration from Eastern Europe has much larger marginal effects of relative unemployment and relative incomes compared to the model for “Western countries”. The long-run relative income elasticity is roughly twice as high and the unemployment term four times as high. In the long run, only relative unemployment rates affect the immigration rate. The estimates show that the immigration rate increased a lot due to the EU enlargements of 2004 and 2007. As we can see from equaiton (9) the t-value for the coefficient of the lagged dependent variable is −6.56 which according to the cointegration tests in Pesaran et al. (2001) clearly rejects that variables in levels should be excluded from our model thus supporting cointegration. The recursive parameter estimates of eq. (9) from 2008 and onwards when both Bulgaria and Rumania had become members of the EU are stable (available upon request). Equation (9) is included in the simulation model used in this paper.
About this article
Cite this article
Cappelen, Å., Eika, T. Immigration and the Dutch disease A counterfactual analysis of the Norwegian resource boom 2004-2013. Open Econ Rev 31, 669–690 (2020). https://doi.org/10.1007/s11079-019-09543-9
- Dutch disease