De Economist (2013) 161:118 DOI 10.

1007/s10645-012-9197-6

The Contribution of Migration to Economic Development in Holland 15701800

Peter Foldvari Bas van Leeuwen Jan Luiten van Zanden

Received: 20 December 2011 / Accepted: 31 August 2012 / Published online: 12 September 2012 Springer Science+Business Media New York 2012

Abstract Migration always played an important role in Dutch society. However, little quantitative evidence on its effect on economic development is known for the period before the twentieth century even though some stories exist about their effect on the Golden Age. Applying a VAR analysis on a new dataset on migration and growth for the period 15701800, we nd that migration had a positive effect on factor accumulation during the whole period, and a positive direct effect on the per capita income during the Golden Age. This seems to conrm those studies that claim that the Dutch economy during its Golden Age at least partially benetted from immigration. Keywords Economic growth Immigration Holland Endogenous development Human capital JEL Classication J15 N13 N33

P. Foldvari B. van Leeuwen (B ) J. L. van Zanden Economic and Social History Department, Utrecht University, Drift 17, Utrecht 3512 BS, The Netherlands e-mail: bas.vanleeuwen1@googlemail.com P. Foldvari Debrecen University, Debrecen, Hungary B. van Leeuwen Warwick University, Coventry, UK J. L. van Zanden Groningen University, Groningen, The Netherlands

123

P. Foldvari et al.

1 Introduction Migration is a hot topic. Historians, not entirely insensitive for those kinds of societal debates, have also turned to examples of large migration ows to study their long term impact on society and economy. The Dutch Republic is one of those cases that have been discussed extensively in this literature. There is consensus that a lot of migration occurred during its Golden Age (15801670). It has also been concluded that those large migration ows did not have negative effects on society and economy (for example Lucassen and Lucassen 2011). The debate we engage with in this paper is about the question how important immigration was for economic success. Here we can distinguish two views. One view, argued most forcefully by Israel (1989) and echoed until today (e.g. Esser 2007), is that the spectacular success of the Dutch republic after 1580 was to a very large extent due to the immigration of highly schooled and relatively wealthy entrepreneurs and skilled labourers from the southFlanders and Brabant. They ed for the Spanish forces, relocated in the cities of Holland and Zeeland, and brought with them the high-valued added activities that created a big economic boost. In other words, the Golden Age mainly consisted of the relocation of the economic centre of the Low Countries from Antwerp and the surrounding areas to Amsterdama process resulting from the Spanish reconquest of the south. In this scenario, the migration ow of the period between the 1580s and the 1620s is the decisive link between Flemish and Dutch prosperity. Other authors (van Zanden 1993; de Vries and van der Woude 1997) have, in contrast, argued that the growth of the Holland economy was rst of all based on indigenous developments: the emergence of an efcient set of institutions there, set in motion a process of autonomous economic growth, which already started between 1350 and 1500 when, for example, the share of urbanisation rose from 23 to 40 % making Holland one of the most urbanised (and non-agricultural) regions in the world. In this view, the growth spurt of the Golden Age was the continuation of a process of economic growth that began much earlier. It was also logical that the expelled merchants and craftsmen of Flanders after 1585 moved to Holland and Zeeland, because this region offered by far the most attractive opportunities for themsuch as an efcient set of institutions. In the endogenous-growth-model the immigration wave of 15801620 is a relevant and important development, but its contribution to long term economic growth is limited. What is perhaps more important in this approach (as formulated by de Vries and van der Woude 1997 and by van Zanden 1993), is that the Holland labour market was a very open one, which, when the economy accelerated after 1580, was able to attract increasingly large numbers of labourers from the rest of the Netherlands (Brabant, Overijssel, Friesland) and from parts of Germany and Scandinavia. The VOC, for example, became an employer of thousands of sailors and soldiers recruited from all parts of the North Sea area. It has been argued that this very exible supply of unskilled and semi-skilled labour, which continued during the seventeenth and eighteenth centuries, was a key to the long-term economic success of the region. The discussion on the links between economic development and migration so far has concentrated on these themes (there are no contributions which approached this

123

The Contribution of Migration

subject from another perspectivelooking at HeckscherOhlin forces, for example). The debate has mostly centred on measuring the numbers of migrants and on their impactbut only few have attempted to quantify that impact (but see Gelderblom 2000). It has fortunately become possible to bring more sophisticated statistical methods into the debate because we have just nished a large research project constructing the national accounts of Holland on an annual basis between 1514 and 1807 (in fact, the series goes back to 1347, but the pre 1514 estimates are very tentative). Moreover, we are now also able to estimate the inow of migrants into Holland somewhat better, thanks to new estimates of the demographic development of Holland between 1514 and 1807, a spinoff of the project on reconstructing the national accounts. As a result, it is now, for the rst time, possible to test the ideas on the relation between migration and economic growth more rigorously.

2 Data The main datasets on the economy of Holland used here have been introduced and explained in detail in other papers. The focus is on Holland, the biggest province in the Netherlands in the early modern period, approximately equal to the current provinces of Northern -and Southern Holland. It was also the most dynamic and richest part of the early modern period and, hence, the region that profited most from the Golden Age. The main results of the project on the reconstruction of the national accounts of Holland in the period before 1800 have been presented in van Zanden and van Leeuwen (2012), where it is explained how the estimates of GDP, GDP per capita and population have been constructed. For the period after 1514, estimates of total GDP were the result of putting together value added series for 27 branches of the economy (from agriculture to banking); the evidence for the 13471514 period is much weaker, but we will not include this period into the analysis of this paper. Moreover, using a method developed by Feinstein and Thomas (2001), we were also able to estimate the margins of error of the GDP gures. Figures 1 and 2 printed below report the main ndings. The estimates demonstrate that the period 15801670the classical Golden Agewas a period of rapid growth of total GDP and of the population of Holland, but in terms of intensive growththe growth of GDP per capitait was not exceptional. As Fig. 1 shows, there was already strong growth of GDP per capita in the late medieval period (but the margins of error of these estimates are quite large). It also is clear that this trade-oriented economy was characterized by a relatively high level of instability of GDPmainly due to exogenous shocks (wars, harvest failures etc.). Our estimates are also rather positive about growth during the eighteenth century, which has often been portrayed as a period of economic stagnation. We nd continuous per capita growth during that century, albeit that total GDP and total population is growing at a much slower pace. The Golden Age is therefore in the rst place a period of very rapid population growth, whereas the pace of intensive growth seems to be rather stableboth before and after the seventeenth century. It is thus clear that, the period 15741650 saw considerable growth in both per capita output and population. Immigration was a main factor behind this sharp increase in

123

P. Foldvari et al.

Fig. 1 Per capita GDP (1,800 constant guilders, including error margins). Source van Zanden and van Leeuwen (2012)

Fig. 2 GDP (million 1990 GK dollars), population (*1,000) in Holland, 13471807. Source van Zanden and van Leeuwen (2012)

population growth during the post 1580 period. In another paper we have presented estimates of the main demographic features of the Holland, including estimates of the minimum level of immigration to the region. The total population of Holland increased from 275,000 in 1514 to 400,000 in 1572an increase that was almost entirely the result of its own natural increase. After 1572 there was rst a small dip, followed by very rapid growth resulting in a peak level of about 880,000 in 1672. This was followed by a moderate decline to about 783,000 in 1750, after which the population stabilized at this level for about 50 years. This stabilization remained until the mid-nineteenth century. Afterwards, we saw larger number of migrants entering the Netherlands, but never in those magnitudes as recorded in the seventeenth century. Figure 3 presents the estimates of the population curve of Holland, including our estimates of net immigration. In the nal decades of the sixteenth century net immigration (from outside of Holland) was about 3,800 per year, to increase to on average 5,200 during the seventeenth century; the peak of around 10,000 immigrants occurred about 1650. Total immigration in Holland between 1574 and 1650 is estimated at

123

The Contribution of Migration

Fig. 3 Population (in 1,000, right-hand scale); Annual number of immigrants (in 1,000, Left-hand scale) for Holland and the Netherlands, 15101800. Source This paper, Oomens (1989), and 200 jaar statistiek in tijdreeksen

480,000so larger, for example, than the original population of Holland in 1570. These are lower bound estimates; they are based on the difference between the natural increase of the population and its actual growth, and therefore do not include people who emigrated from Holland (for example, left on the ships of the East Indies Company); their total number is roughly estimated at about 200250,000 (Lucassen 2002), bringing total immigration to about 700,000. We also ignore in our estimates temporary migratory ows, such as the seasonal workers analysed by Lucassen (1987). Clearly, immigration was huge in the late sixteenth and seventeenth century. In the late seventeenth and eighteenth century the number of migrants fell to on average 1,300 migrants per year while in nineteenth century the number of migrants increased from ca. 7,000 per annum in the 1860s to roughly 17,000 in the 1890s. Even though these latter migrants were bigger in number than in the Golden Age, we have to be aware that they made up a far smaller proportion of the total population. Lucassen (2002) and Oomens (1989), for example, calculated that, whereas the share of migrants in the population in the 1890s was around 1.6 %, in 1600 it was no less than 10 %. And these numbers are for the Netherlands, while most migrants would have travelled to Holland. Migration can have a direct on economic growth (for example via technological development) but it may also work via the factors of production such as physical- or human capital if the migrants brought these two assets along with them to Holland/the Netherlands. Therefore, in our following analysis we also include series of physicaland human capital. We will use both the human capital (i.e. average years of education in the population aged 15 and older) and non-residential physical capital for Holland (van Zanden and van Leeuwen 2012). In Table 1 we report the unit root test of our 4 variables: log of real GDP per capita (lny), physical capital per capita (lnk), average years of education (avyears), and number of migrants per 1000 inhabitants (migration) for the sub-periods 1572 1650, 16501700 and 17001800. The sub-periods follow the standard periodization

123

6 Table 1 Unit root tests 15721650 ADF lny lnk migration avyears lny lnk migration avyears 5.051 4.930 3.712 2.345 9.701 5.184 5.930 2.767 KPSS 0.130 0.069 0.167 0.251 0.166 0.066 0.029 0.157 16501700 ADF 3.889 2.034 1.767 2.398 8.902 5.001 9.222 3.410 KPSS 0.169 0.135 0.188 0.175 0.500 0.162 0.479 0.180

P. Foldvari et al.

17001800 ADF 2.772 2.078 2.600 1.055 11.734 2.997 8.886 2.861 KPSS 0.109 0.155 0.093 0.093 0.055 0.217 0.035 0.447

For the levels we used a test specication with constant and trend, for the differences we used a specication with constant only. The null-hypotheses of the ADF and KPSS tests are non-stationarity and stationarity respectively. For the ADF tests we used an automatic lag selection (max. lag = 20 with the SBC as model selection criterion). For the KPSS test we used Bartlett kernel with automatic NeweyWest bandwith choice

of the economic development of the Dutch Republic with 15721650 being the Golden Age, 16501700 a period of crisis and 17001800 a period of stagnation. We adopt the same periodization for the VAR analysis as well. We intentionally used two unit-root tests with different null-hypotheses: while the ADF has the null of non-stationarity, the KPSS tests stationarity against the alternative of non-stationarity. The two test sometimes lead to contradicting results: the log of GDP per capita is usually found to be I(1) but for the period of 16501700 the KPSS suggest trend-stationarity while the ADF indicates I(1). Similarly the type of stationarity is not easy to determine for migration: depending on which test we prefer it can be trend-stationary for 15721650 but based on the KPSS test it is rather I(1). In the next section we carry out Johansen cointegration tests and also discuss the effect of migration on per capita real GDP.

3 Empirical Analysis In this paper we aim at estimating the effect of migration on economic growth. In the literature, it is argued that migration can have a direct impact on economic growth, or indirect via the factors of production (e.g. Dolado et al. 1994; Walz 1995).1 Likewise, Morley (2006) argues for a reverse causality between migration and growth. We will rely on VAR system to draw conclusions about the direction of causality and the existence of a long-run relationship (cointegration) and use impulse response functions (IRF) to obtain a picture of the dynamics of the relationships and to estimate its long-run effect.
1 Just to mention some examples: direct effects may rise due to an increased demand for goods and housing,

while indirect effects may result from different propensities to save or different attitudes toward education that affect physical- and human capital accumulation in the long-run.

123

The Contribution of Migration

3.1 Granger Causality Tests In order to identify the causal relation between the different variables, we start with a Granger causality test. Variable X is said to Granger-cause variable Y if the past values of X contain useful information on the current value of Y (Granger 1969). The standard procedure involves tting the best possible autoregressive model on the values of Y and introducing past values of X as additional explanatory variables to the specication. If a joint significance test of the coefcients of the lags of X suggests that they actually improved the t of the model, we can reject the null-hypothesis of the lack of Granger-causality. Possible pitfalls of this methodology may include omitted variable bias, incorrect choice of the number of lags, and also the effect of non-stationary variables. Obviously, a VAR-system is ideally suited for a Granger-test on stationary variables, but as Toda and Yamamoto (1995) claim, standard Granger-causality tests can be misleading in the presence of integrated series. Using a Granger-causality test on a VEC (Vector Error Correction) system or on a VAR on differenced variables, however, would also be misleading since taking rst differences would remove the possible long-run relationship among the endogenous variables. Additionally, unnecessary differencing may increase the error to signal variance ratio in the presence of measurement errors (Plosser and Schwert 1978). Therefore, Toda and Yamamoto suggest a procedure that enables the test for Granger-causality even in the presence of integrated variables and cointegration.2 First, one identies the highest order of integration (denoted as m) in the endogenous variables and estimates the best possible VAR(p) model. The second step is a Granger-causality test that should be carried out on the rst p lags of a VAR(p + m) system. Since we found that the highest order of integration was one for all periods we use the m = 1 assumption. Our decision regarding the lag length of the VAR system (p) is not solely based on model selection criteria; if at the suggested lag length we still nd residual autocorrelation significant at 5 % we add further lags as long as it disappears. Also if the stability conditions were not fullled for the VAR system, we increase the order of the VAR as long as we obtain no characteristic roots outside the unit circle.3 This strategy sometimes leads to high order VAR systems. The results of the specication process are summarized in Table 2 while Table 3 contains the results from the TodaYamamoto Granger-causality test. For all the periods we nd evidence that migration Granger caused some macroeconomic variable of interest: for the pre-1700 period we nd no direct effect of migration on GDP per capita, but we do nd a causality running from migration toward physical capital stock. For the period 15721650 we nd that physical capital Granger caused per capita GDP which suggests the existence of an indirect link through which migration may have affected per capita income.
2 As unit-root tests have generally low power and the results from the cointegration tests may be sensitive

to the choice of lags or affected by the measurement error in our data, we decide to follow Toda and Yamamotos method as it allows carrying out a Granger test without transforming the model based on possibly biased test results.
3 Thereby we assure that the VAR is invertible to a VMA representation and we can obtain meaningful

impulse response functions.

123

8 Table 2 VAR specications 15721650 Lag length preferred by FPE Lag length preferred by AIC Lag length preferred by SBC Our choice (p) 3 10+ 2 11 16501700 10+ 10+ 2 7

P. Foldvari et al.

17001800 3 3 2 10

FPE nal prediction error, AIC akaike information criterion, SBC Schwarz Bayesian information criterion, we choose the lag length (p) so that no residual autocorrelation signicant at 5 % remains and the system fulls that stability conditions Table 3 Results of the TodaYamamoto (1995) Granger test (only p values are reported) Period Explanatory variables Dependent variables log GDP p.c. log capital p.c. 0.335 0.070 0.345 0.481 0.000 0.122 0.031 0.212 0.033 Migration Av. years of schooling 0.152 0.698 0.059 0.025 0.119 0.390 0.024 0.268 0.684

15721650

log GDP p.c. log capital p.c. Migration Av. years of schooling

0.008 0.181 0.638 0.268 0.723 0.179 0.020 0.008 0.265

0.838 0.799 0.010 0.012 0.024 0.000 0.840 0.802 0.438

16501700

log GDP p.c. log capital p.c. Migration Av. years of schooling

17001800

log GDP p.c. log capital p.c. Migration Av. years of schooling

Bold values are causality signicant at 5 %, i.e. a p value less than 0.1 (0.05) means that the variable in the respective row Granger caused the variable in the respective column at 10 % (5 %) level of signicance

3.2 Cointegration Analysis For a possible existence of long-run relationship among the variables we apply Johansen cointegration tests on the above estimated VAR specications. This test is based on a Vector Error-Correction representation of the processes.
p 1

Yt = +
i =1

Yt i +

Yt 1 + et

where the rank of matrix (r( )) is indicative of the existence of cointegration and the number of cointegrating vectors. If there are k endogenous variables, if 0 < r ( ) < k, then r( ) equals the number of cointegrating vectors. If r( ) = 0 then the variables are non-stationary but not cointegrated. If r( ) = k the variables are stationary and

123

The Contribution of Migration Table 4 Results from the trace and maximum eigenvalue tests (only p values are reported) Rank of matrix Trace test At most 0 At most 1 At most 2 At most 3 Maximum eigenvalue test At most 0 At most 1 At most 2 At most 3 Suggested rank (at 10 % sign.) 0.000 0.000 0.060 0.431 3 0.000 0.089 0.024 0.068 4 0.000 0.026 0.028 0.013 4 0.000 0.000 0.074 0.431 0.000 0.035 0.012 0.068 0.000 0.000 0.004 0.013 15721650 16501700

17001800

Trace test for rank k: H0 rank is at most k, H1 rank is 4 Maximum eigenvalue test for rank k: H0 rank is at most k, H1 rank is k + 1

by definition cannot be cointegrated. The standard testing methods include the trace and the maximum eigenvalue test. Table 4 has the outcomes: We nd that with the exception of 15721650 matrix is of full rank, indicating that the variables were stationary even though this contradicts some of the unit-root tests results reported in Table 1. Still, unit-root tests generally have low power so we prefer the results from the Johansen-test. It should be noted that the results from the Johansen test of cointegration is sensitive to the choice of lag. Generally if the order of the VAR system is chosen too low, the test has the tendency to nd spurious cointegration (Cheung and Lai 1993).4 At 10 % level of significance we nd evidence for three cointegrating vectors for the Golden Age and nd the matrix of full rank for the rest of the periods meaning that the variables should be I(0).5

4 In other words, if we had accepted the suggestion by the Schwarz Information Criterion and had estimated VAR(2) or VAR(3) systems for all sub-periods, we would have found one cointegrating vector for 1572 1650, and no cointegrating vectors for 16501700 and 17001800 with the variables being non-stationary (that is the Johansen test could not reject that the rank of matrix was zero. The presence of residual autocorrelation in those models, however, is a clear warning that these specications could not completely capture the dynamic of the variables. Furthermore, Cheung and Lai (1993) nd that the Johansen test is sensitive to underparametrization (choosing too few lags) and the results can be biased toward nding spurious cointegrating vectors. They claim that Johansen test is robust to the overparametrization, however, hence we rather take the risk of overtting the model at the price of losing some efciency than undertting it. The obtained CIRFs would only be qualitatively different in case of 15721650 where the CRIFs from a VEC(1) specication would reect a permanent effect of migration on all endogenous variable. This result does not make any sense as with the observed inow of immigrant we should observe an accelerating growth of per capita income in the period which is obviously not found in the data. 5 We opted to decide based on a 10 % level of significance because due to the limited sample sizes (79, 51 and 101 years respectively) and the presence of measurement error in historical estimates. With 1 % level of significance we would nd 3 cointegrating vectors for all the periods.

123

10 Table 5 Restricted cointegrating vector for 15721650 CV 1 Constant lny lnk migration avyears 6.129 1 0.861 (1.296) 0 0 0.012 (4.03) 0 The LR test for binding restrictions p value 0.144 CV2 2.506 0.312 (3.056) 1

P. Foldvari et al.

CV3 0.774 0 0 0.044 (4.88) 1

Table 6 Restricted adjustment coefcients for 15721650 CV 1 lny lnk migration avyears 0 0.121 (2.081) 0 0 CV2 2.783 (4.055) 0.272 (2.676) 0 0.026 (1.734) 6.656 (2.596) 0.035 (4.314) 0 CV3 0

For the 15721650 period we tested different restrictions on the cointegrating vector so that we get a better insight to the long-run relationships. The restricted cointegrating vector, with the adjustment coefcients are reported in Tables 5 and 6. We choose the coefcients of the log GDP per capita, log capital per capita and the average years of education to be normalized to unit respectively. Neither migration nor average years of education were found to yield a significant long-run coefcient in the rst cointegrating vector, so they were omitted. This means that there was no direct long-run relationship among the per capita GDP and the migration. On the other hand we nd evidence for indirect relationships: rst, migration seems to have had a positive relationship with per capita physical capital stock (second cointegrating vector) and also with average years of education (third cointegrating vector). This leads to the conclusion that immigrants during the Golden Age did not necessarily contribute to higher productivity in Holland, but rather brought a different attitude to factor accumulation, with higher propensity to save and invest and higher likelihood to follow some formal education. These attitudes are expected to have been benecial for the rise of commercial capitalism.

123

The Contribution of Migration

11

3.3 Impulse-Response Functions The estimation of impulse-response functions can be a way to obtain a better understanding of the dynamic relationship among the endogenous variables, and to have an estimate on the long-run effect as well. Still, simply estimating IRFs on the baseline VAR model would be likely to lead to biased estimates and wrong conclusions if the variables are in simultaneous relation (contemporaneously correlated). In order to see why this is the case, let us have the following VAR(p) system:
p

AYt = 0 +
i =1

i Yt i + ut , YtT = (ln yt , ln kt , migration t , a v year st )

Where matrix A has the coefcients of the simultaneous relationship among the Y variables. Obviously if A is an identity matrix then the variables are not correlated contemporaneously and the IRFs on the baseline model can be trusted. If this is not the case, however, the residuals form the VAR will contain not only the shocks to a given variable, but also the effect of innovations in other variables, or in other words, the residuals will be correlated:
p

Yt = A1 0 +
i =1

A1 i Yt i + A1 ut

So before any meaningful IRF can be estimated from this model, one needs to have certain assumptions about matrix A, which involves a structural factorization (estimation of a Structural VAR or SVAR). A useful check for the existence of a simultaneous relationship among our endogenous variables is to check if there is some linear correlation among the VAR residuals. The results are included as Table 7. As for 15721650 and 16501700 we nd only two possible simultaneous relationships: one is between average years of education and log of GDP per capita, the other is between average years of education and migration. For 17001800 we obtain a significant correlation coefcient for the residuals of the log capital stock and log of GDP per capita, and the average years of education and migration. The identication of the matrix A requires that the correlation is attributed to only one of the variables. We operate on the assumption that it is more likely that the GDP per capita was affected by average years of education, and not vice versa, and migration affected education, so the observed correlation can be attributed fully to migration. For the period 1700 1800 we assume that a shock in GDP per capita had an immediate impact on capital stock, but not vice versa.6 The IRFs and the cumulated IRFs are reported as Figs. 4, 5, 6, and 7. The impulse response functions reveal that migration had a positive level effect on GDP per capita during the Golden Age, while we nd a negative effect for

6 The obtained IRFs are not much different without a structural factorization either.

123

12 Table 7 Correlation matrices of the VAR residuals lny 15721650 lny lnk migration avyears 16501700 lny lnk migration avyears 17001800 lny lnk migration avyears 1.000 0.283 0.010 0.050 0.283 1.000 0.146 0.055 0.010 0.146 1.000 0.357 1.000 0.190 0.189 0.430 0.190 1.000 0.071 0.087 0.189 0.071 1.000 0.283 1.000 0.021 0.043 0.239 0.021 1.000 0.067 0.086 0.043 0.067 1.000 0.856 lnk migration

P. Foldvari et al.

avyears 0.239 0.086 0.856 1.000 0.430 0.087 0.283 1.000 0.050 0.055 0.357 1.000

denotes that a linear correlation coefcient is signicant at at least 10 %

17001800. Whether these effects are statistically significant is disputable. We report some analytic condence intervals for the short run-multipliers and these often include zero as well. We should not forget however that as with historical data generally is the case, an increased degree of uncertainty comes from the increased measurement errors. Yet, even with this, the effect of migration on capital stock seems to be significant by all means, so even if one remains at the safe side, the nding that migrants positively affected capital accumulation in all sub-periods seems quite robust. The long-run effects have been estimated from the cumulated IRFs. The CIRFs seem to converge to a value after 20 years so it is safe to take their value at 20 years as an estimate of the long-run effect. These estimates are reported as Table 8.

4 Conclusion Migration always played an important role in Dutch society. The absolute number of migrants clearly increased strongly during the 80 years war and the Golden Age. However, as argued by Esser (2007, 268), this was not so much a break with the past but rather an acceleration of already existing migrant ows. We do nd, however, that migrant ows dropped from the Golden Age to the eighteenth century, to increase again thereafter. Yet, even though the absolute number of migrants was highest in the late nineteenth century, relative to the population their numbers were much more significant during the Golden Age.

123

The Contribution of Migration

Response of log GDP p. c
.12 .08 .04 .00 -.04 -.08 2 4 6 8 10 12 14 16 18 20 .03 .02 .01 .00 -.01 -.02 2 4 6 8 10 12 14 16 18 20

13
Response of log capital stock p. c

Response of migration
1.0 0.5 0.0 -0.5 -1.0 -1.5 2 4 6 8 10 12 14 16 18 20 .020 .015 .010 .005 .000 -.005

Response of av. years of education

10

12

14

16

18

20

Accumulated Response of log GDP p.c.

.20 .15 .10 .05 .00 -.05 -.10 2 4 6 8 10 12 14 16 18 20 .04 .00 .16 .12 .08

Accumulated Response of log capital stock p.c

-.04 2 4 6 8 10 12 14 16 18 20

Accumulated Response of migration

5 4 3 2 1 0 -1 -2 2 4 6 8 10 12 14 16 18 20

Accumulated Response of av. years of schooling

.20 .15 .10 .05 .00 -.05 2 4 6 8 10 12 14 16 18 20

Fig. 4 Impulse response function 15721650 based on SVAR(11), responses to one SD (3,040 immigrants) impulse in migration (2 SE condence intervals)

We applied a VAR system on a newly available dataset to draw conclusions about the causality and long-run relationships of migration and other macro-economic variables. Interestingly during the Golden Age migration had a positive long-run direct effect on GDP per capita. It also positively affected capital accumulation and the level

123

14
Response of log GDP p.c.
.12 .08 .04 .00 -.04 -.08 2 4 6 8 10 12 14 16 18 20 .03 .02 .01 .00 -.01 -.02 -.03 2 4 6 8 10 12 14

P. Foldvari et al.
Response of log capital stock p.c.

16

18

20

Response of migration
1.5 1.0 0.5 0.0 -0.5 -1.0 2 4 6 8 10 12 14 16 18 20 .020 .016 .012 .008 .004 .000

Response of av. years of education

10

12

14

16

18

20

Accumulated Response of log GDP p.c.

.24 .20 .16 .12 .08 .04 .00 -.04 -.08 2 4 6 8 10 12 14 16 18 20

Accumulated Response of log capital stock p.c.

.25 .20 .15 .10 .05 .00 2 4 6 8 10 12 14 16 18 20

Accumulated Response of migration

5 4 3 2 1 2 4 6 8 10 12 14 16 18 20

Accumulated Response of av. years of education

.16 .12 .08 .04 .00 2 4 6 8 10 12 14 16 18 20

Fig. 5 Impulse response function 15721650 based on VEC(10), responses to one SD (3,040 immigrants)impulse in migration (no condence intervals are available)

of education in the population. This changed after 1650 when the effect of migrants on economic development either directly or via the factors of production became insignificant. After 1700 the positive effect on physical capital went up again, but the

123

The Contribution of Migration

Response of log GDP p.c.
.04 .02 .00 .000 -.02 -.04 2 4 6 8 10 12 14 16 18 20 -.004 -.008 2 4 6 8 10 12 14 16 18 20 .012 .008 .004

15
Response of log of capital stock p.c.

Response of migration
2 1 0 -.002 -1 -2 2 4 6 8 10 12 14 16 18 20 -.004 -.006 -.008 .006 .004 .002 .000

Response of av. years of schooling

10

12

14

16

18

20

Accumulated Response of log GDP p.c.

.10 .05 .00 -.05 -.10 -.15 2 4 6 8 10 12 14 16 18 20

Accumulated Response of log of capital stock p.c.

.05 .04 .03 .02 .01 .00 -.01 -.02 2 4 6 8 10 12 14 16 18 20

Accumulated Response of migration

8 6 4 2 0 -2 -4 2 4 6 8 10 12 14 16 18 20

Accumulated Response of av. years of education

.04 .02 .00 -.02 -.04 -.06 2 4 6 8 10 12 14 16 18 20

Fig. 6 Impulse response functions 16501700, responses to one SD (5,270 immigrants) impulse in migration (2 SE condence intervals)

direct effect on GDP per capita became negative, hence, cancelling each other out to a certain extent. This means that only during the Golden Age the net effect of migration on per capita GDP was positive and significant which conrms those studies that claim that, for example, rich merchants went to Amsterdam and brought their capital and networks along (van Dillen 1958; Brulez 1960). Altogether, the positive effect

123

16
Response of log GDP p.c.
.03 .02 .01 .00 -.01 -.02 -.03 2 4 6 8 10 12 14 16 18 20 .000 -.004 -.008 2 4 6 8 10 12 14 .012 .008 .004

P. Foldvari et al.
Response of log of capital stock p.c.

16

18

20

Response of migration
.6 .4 .2 .0 -.2 -.4 2 4 6 8 10 12 14 16 18 20 .003 .002 .001 .000 -.001 -.002

Response of av. years of education

10

12

14

16

18

20

Accumulated Response of log GDP p.c.

.10 .05 .00 -.05 -.10 -.15 -.20 2 4 6 8 10 12 14 16 18 20

Accumulated Response of log capital stock p.c.

.16 .12 .08 .04 .00 -.04 2 4 6 8 10 12 14 16 18 20

Accumulated Response of migration

4 3 2

Accumulated Response of av. years of education

.03 .02 .01 .00

1 0 2 4 6 8 10 12 14 16 18 20

-.01 -.02 2 4 6 8 10 12 14 16 18 20

Fig. 7 Impulse response function 17001800, responses to one SD (1,191 immigrants) impulse in migration (2 SE condence intervals)

123

The Contribution of Migration

17

Table 8 Estimated long-run effect of 1,000 new immigrants after 20 years (based on the cumulated IRFs) 15721650 SVAR(11) On per capita GDP (%) On per capita capital stock (%) On migration On average years of education (years) 2.1 2.0 621 0.028 15721650 VEC(10) 3.0 2.8 913 0.003 16501700 SVAR(7) 0.5 0.4 224 0.001 17001800 SVAR(10) 4.1 5.4 1,517 0.005

For the period 15721650 we report the long-run effects from a VEC(10) specication as well, but the results are subject to the assumption of no contemporary correlation of the variables. This can cause a bias

of migrants on factor accumulation is strong indication of their significant role in the success of commercial capitalism in Holland during the Golden Age. References
Brulez, W. (1960). De diaspora der Antwerpse kooplui op het einde van de 16e eeuw. Bijdragen Voor Geschiedenis der Nederlanden, 15, 229306. Cheung, Y.-W., & Lai, K. S. (1993). Finite sample sizes of Johansens likelihood ratio tests for cointegration. Oxford Bulletin of Economics and Statistics, 55(3), 313328. de Vries, J., & van der Woude, A. (1997). The rst modern economy success failure and perseverance of the Dutch economy. Cambridge: Cambridge University Press. Dolado, J., Goria, A., & Ichino, A. (1994). Immigration, human capital and growth in the host country: Evidence from pooled country data. Journal of Population Economics, 7(2), 193215. Esser, R. (2007). From province to nation immigration in the Dutch Republic in the late 16th and early 17th centuries. In S. G. Ellis & L. Klusakova (Eds.), Imaging frontiers, contesting identities (pp. 163276). Pisa: Pisa University Press. Feinstein, C. H., & Thomas, M. (2001). A plea for errors. Discussion papers in economic and social history, number 41, University of Oxford. Gelderblom, O. (2000). Zuid-Nederlandse kooplieden en de opkomst van de Amsterdamse stapelmarkt. Hilversum: Verloren. Granger, C. W. J. (1969). Investigating causal relations by econometric models and cross-spectral methods. Econometrica, 37(3), 424438. Israel, J. (1989). Dutch primacy in world trade 1989 (pp. 15851740). Oxford: Oxford University Press. Lucassen, J. (1987). Migrant labour in Europe: The drift to the North Sea. Beckenham: Croom Helm. Lucassen, J. (2002). Immigranten in Holland, Een kwantitatieve benadering. IISH working paper series no. 3. Lucassen, L., & Lucassen, J. (2011). Winnaars en Verliezers. Amsterdam: Bert Bakker. Morley, B. (2006). Causality between economic growth and immigration: An ARDL bounds testing approach. Economics Letters, 90(1), 7276. Oomens, C. A. (1989). De loop der bevolking van Nederland in de negentiende eeuw Statistische onderzoekingen nr M35. Den Haag: SDU-uitgeverij/CBS-publicaties. Plosser, N., & Schwert, C. (1978). Money income and sunspots measuring economic relationships and the effects of differencing. Journal of Monetary Economics, 4, 647660. Toda, H. Y., & Yamamoto, T. (1995). Statistical inference in vector autoregressions with possibly integrated processes. Journal of Econometrics, 66(12), 225250. van Dillen, J. A. (1958). Het oudste aandeelhoudersregister van de kamer Amsterdam der Oost-Indische Compagnie. Martinus Nijhoff: The Hague. van Zanden, J. L. (1993). The rise and decline of Hollands economy. Manchester: Manchester University Press.

123

18

P. Foldvari et al.

van Zanden, J. L., & van Leeuwen, B. (2012). Persistent but not consistent: The growth of national income in Holland 13471807. Explorations in Economic History, 49(2), 119130. Walz, U. (1995). Growth (rate) effects of migration. Zeitschrift fr Wirtschafts- Und Sozialwissenschaften, 115, 199221.

123