Economics of Education

Transcription

1 Economics of Education Munich, 1 2 September 2017 Breaking the Links: Natural Resource Booms and Intergenerational Mobility Aline Bütikofer, Antonio Dalla-Zuanna, and Kjell G. Salvanes

2 Breaking the Links: Natural Resource Booms and Intergenerational Mobility Aline Bütikofer Norwegian School of Economics Antonio Dalla-Zuanna Norwegian School of Economics Kjell G. Salvanes Norwegian School of Economics July 21, 2017 Abstract Do large economic shocks increase intergenerational earnings mobility by creating new economic opportunities or do they instead reduce mobility by reinforcing the links between generations? In answering this question, we estimate how the Norwegian oil boom starting in the 1970s affected intergenerational mobility in those local labor markets most affected by the growing oil industry. We find that this resource shock increased intergenerational mobility for cohorts commencing their professional careers at the beginning of the oil boom. Importantly, these findings are not driven by preexisting local level differences in intergenerational mobility or regional differences in education, nor are they sensitive to selective migration or adverse health effects. Instead, the change in intergenerational mobility is mostly driven by bottom-up mobility and a decrease in the returns to academic education in oil-affected regions. The findings also persist across a third generation, with intergenerational mobility being significantly higher for boom-affected areas in both grandfather son and father son comparisons. 1 Introduction During recent decades, many societies have experienced sharp increases in economic and social inequalities. An important issue is whether these increases in inequality persist across generations We gratefully acknowledge comments by Nathaniel Hendren, Jan Stuhler, Bentley MacLeod, Janet Currie, José V. Rodríguez Mora, Sandra E. Black, Paul Devereux, Martin Nybom, and seminar, workshop, and conference participants at the University of Melbourne, the University of Sydney, the Frisch Center, the University of Naples Federico II, the CReAM seminar at University College London, the University of Bergen, the Norwegian School of Economics, the Madrid Mobility Workshop 2016, the Nordic Summer Institute in Labor Economics, the SDU Applied Microeconomics Workshop, the 2017 Royal Economic Society Annual Conference, and the 31st Annual Conference of the European Society for Population Economics. Lene Bonesmo Solberg provided excellent research assistance. 1

3 and thereby lower intergenerational mobility. However, while the literature has documented differences in intergenerational mobility across regions within countries and changes over time (Corak, Lindquist, and Mazumder, 2014; Chetty, Hendren, Kline, Saez, and Turner, 2014; Nybom and Stuhler, 2013; Pekkarinen, Salvanes, and Sarvimäki, 2017), the factors that determine changes and regional differences in intergenerational mobility are not well understood. In particular, precisely how major economic shocks or turbulence affect intergenerational mobility and whether these changes persist across generations remains an open question. The direction of the effect of an economic shock on intergenerational mobility is unclear. On the one hand, if new industries with new job opportunities demanding new skills are established, this may decouple the ties between parents and their children s outcomes. Hence, parental income and existing social networks are less-accurate predictors of children s economic outcomes. On the other hand, poorer families may be less able to benefit from new opportunities leading to lower social mobility. Moreover, different types of economic shocks, including economic downturns or upturns, natural resource booms, or technological change, may well reinforce or break the transmission of economic status and the resulting changes may persist for one or multiple generations. In this paper, we focus on the effect of a specific major and long-lasting resource boom on social mobility. Resource booms substantially increase local economic growth and average wage growth despite substantial population migration (Black, McKinnish, and Sanders, 2005a; Allcott and Keniston, forthcoming; Basso, 2016). However, a resource boom also changes the opportunity costs of attending high school or college as more high-paying low-skill jobs become available (Black, McKinnish, and Sanders, 2005b; Cascio and Narayan, 2015; Morissette, Chan, and Lu, 2015). To analyze the effects of a resource boom on the transmission of economic status, we exploit the geographic variation in the impact of the Norwegian oil boom, which transformed the Norwegian economy from being largely based on shipping, logging, fishing, and food production into a globally successful and resource-based economy. We believe the Norwegian oil boom of the 1970s presents an ideal natural experiment for considering the effect of a natural resource shock on intergenerational mobility for three main reasons. First, most indications suggest that oil has been a blessing for Norway because of its high GDP per capita, relatively low unemployment rate, and sizeable government pension fund. Since oil production commenced in 1971, the expansion of oil activities has had far-reaching effects on Norwegian workers and firms. By 2014, the Norwegian oil sector (including oil-related suppliers) provided about 60 percent of Norway s exports and directly or indirectly employed some nine percent of total employment. Hence, if natural resource shocks do alter the transmission of economic status, the Norwegian oil boom should be large enough to measure any possible impact. Second, Norwegian registry data permit us to link the earnings and education of Norwegian parents to those of their children. While we can observe parental earnings before the discovery of oil, the passage of time also allows us to observe the earnings of their children exposed to the 2

4 oil shock as adults, and likewise for the educational outcomes and early career earnings of their grandchildren. Lastly, there are significant differences between Norwegian local labor markets in the size of the oil sector that create geographic variation, which enables us to compare the transmission of economic status across local labor markets affected unevenly by the oil boom. We examine the effect of this natural resource boom on intergenerational mobility using a local labor market strategy, which exploits variation across local labor markets in their exposure. At the beginning of the oil boom, individuals residing in southwest Norway were mostly affected (Løken, 2010). In addition, Brunstad and Dyrstad (1997) present empirical results indicating that there were significant demand effects for petroleum-relevant occupations in areas in close proximity to the first Norwegian oil fields. These new employment opportunities in oil extraction and in the machine and shipping industries, which were major suppliers to the oil industry, may have increased earnings for some individuals entering the labor market in the 1970s in southwest Norway and changed the intergenerational earnings persistence. Following Chetty, Hendren, Kline, and Saez (2014), we use rank-rank regression models to analyze the impact of the resource shock on intergenerational income mobility. In addition, we use absolute measures of intergenerational mobility to analyze whether these changes are driven by bottom-up or top-down movements. In our main analysis, we focus on cohorts (and their parents) entering the labor market during the first decade of oil extraction in Norway to abstract from any oil shock-biased human capital investment decisions (see, e.g., Cascio and Narayan, 2015). To evaluate whether the impact of the oil boom on intergenerational mobility was causal and not driven by preexisting differences in intergenerational mobility across local labor markets, we also consider the intergenerational mobility of cohorts (and their parents) that were about 40 years of age at the beginning of the oil boom and thereby less affected in their career decisions by the growing oil sector. In addition, we analyze whether the changes in the intergenerational transmission of economic status persist over multiple generations. Overall, we find that the Norwegian oil boom increased intergenerational earnings mobility. Sons born in local labor markets that benefitted most from the oil boom of the 1970s experienced more intergenerational earnings mobility than elsewhere. We find these effects are not driven by preexisting locational differences in intergenerational earnings mobility. The transmission of economic status for cohorts that did not benefit from the oil boom early in their working careers does not differ across high- and low-oil regions and the oil boom increased bottom-up mobility and barely affected individuals growing up in wealthier families. Moreover, we obtain empirical evidence that geographic differences in human capital investment do not provide the mechanism underpinning our findings. However, we do show that the returns to academic education are significantly lower in regions affected by the oil boom, and that this may yield an explanation for our main findings. Moreover, we find that the lower part of the earnings distribution in oil-affected regions shifted 3

5 to the right and that this shift was mostly driven by vocationally trained workers in the oil sector. The results are not sensitive to selective migration, to the exclusion of the Oslo labor market, or to adverse health effects. Including a third generation into our analysis, we find that persistence decreases over generations in both high- and low-oil regions and that intergenerational mobility is significantly higher in high-oil regions in both grandfather son and father son comparisons. Although the sons of bottom-to-top movers in the oil-affected regions on average displayed the highest early career earnings, their education level is significantly lower; therefore, they are more vulnerable to negative oil shocks and lower returns to vocational education. Our findings contribute to the very scarce literature establishing a link between macroeconomic conditions and the transmission of economic status. To our knowledge, Feigenbaum (2015) was the first to show that a very different type of economic turbulence in the form of the Great Depression lowered intergenerational mobility in the US for sons growing up in cities hit hardest by economic downturn. In addition, he shows that sons growing up in wealthy families moved to less-affected regions. Unlike Feigenbaum (2015), we consider a boom not a bust followed by an extended period of economic growth, which began in the early 1970s and lasted for more than 40 years. In addition, we contribute to the literature on the dynamics of intergenerational mobility across cohorts (Nybom and Stuhler, 2013) along with the growing literature documenting the substantial geographic variation in intergenerational mobility (Chetty, Hendren, Kline, and Saez, 2014; Chetty and Hendren, 2017). The remainder of the paper is structured as follows. Section 2 provides some historical background on the Norwegian oil boom. Section 3 outlines the empirical strategy and Section 4 discusses the data and some descriptive statistics. Section 5 details the results and Section 6 some robustness tests. Section 7 concludes. 2 Norwegian Oil Exploration and Industry In the late 1950s, few believed that the Norwegian continental shelf concealed rich oil and gas deposits. Fifty years later, the oil and gas industry is now the country s most important industry in terms of both treasury revenue and investment (Ekeland, 2015). While geological opinions were largely negative concerning the presence of oil and gas deposits in the Norwegian parts of the North Sea, the discovery of gas at Groningen in the Netherlands in 1959 revised expectations (Cooper and Gaskell, 1976). In 1963, the Norwegian government proclaimed sovereignty over the Norwegian continental shelf and started issuing licenses to oil companies to carry out preparatory exploration and perform seismic surveys. Drilling commenced in 1965 after agreement on how to divide the continental shelf between Norway, Denmark, and the United Kingdom was reached (see, e.g., Noreng, 1980). In 1965, the Norwegian state issued 22 production licenses for 78 blocks around the southwestern tip of Norway (see, e.g., Helle, 1984). These production licenses provided exclusive rights for 4

6 exploring, drilling, and production in the license area (see Figure A1 in the Appendix). However, as Norway lacked knowledge of platform construction in the 1960s, the first oil rig was towed from New Orleans to Norway in 1966 to drill the first well about 180 kilometers southwest of the Norwegian city of Stavanger. After reaching a depth of 3,015 meters in 84 days, the well failed to find traces of oil and gas. Only the subsequent discovery of Ekofisk, one of the largest offshore oil fields ever found, on December 23, 1969, marked the start of a number of major discoveries on the Norwegian continental shelf (for the location of the Ekofisk field, see Figure A2 in the Appendix). Production from Ekofisk commenced in 1971 (see, e.g., Helle, 1984). In 1972, the Norwegian parliament voted to increase regulations for oil exploration and to develop new knowledge and industries based on petroleum (Finansdepartementet, 1974). Statoil, a state-owned oil company, was funded to look after the government s commercial interests and pursue appropriate collaborations with domestic and foreign oil interests. In addition, the newly established Norwegian Petroleum Directorate was made responsible for recommending which licenses the government should award and ensuring that companies complied with the safety regulations for offshore drilling and production. Hence, 1972 marks a turning point in Norway s petroleum industry. Before 1972, the industry was dominated by foreign oil companies, but the government declared its interest in building up domestic oil expertise in that year (see, e.g., Noreng, 1980). Moreover, the government decided that petroleum from the Norwegian continental shelf must only be landed in Norway (NOU, 1972). 1 These new laws and the establishment of Statoil s headquarters in Stavanger in 1972 transformed what was once a small canning industry town to the new oil capital of Norway. Figure A2 in the Appendix shows that Stavanger was the closest of the three main Norwegian cities (Oslo, Bergen, and Stavanger) to the oil fields, and that international oil companies had already constructed supply bases close to Stavanger in both Tananger and Dusavik before the new law was enacted. Hence, individuals residing in southwestern Norway around Stavanger were mostly affected by the growing oil industry in the 1970s (see, e.g., Løken, 2010) as the oil boom created a substantial labor demand shock, mostly for skilled and semiskilled craftsmen (Brunstad and Dyrstad, 1997). Following a continuing series of large oil and gas discoveries on the Norwegian continental shelf in the North Sea (southwest of Norway), new oil and gas fields were discovered in the Norwegian Sea (off mid-norway) in 1981 and the Barents Sea (northern Norway) in 1984 (see, e.g., Lerøen, 1990). The last large oil discovery was the Johan Sverdrup field in 2010 in the North Sea, which is currently Norway s second-largest field in terms of remaining reserves. Figure A2 in the Appendix displays all oil discoveries until As oil and gas were discovered in the Norwegian and the Barents Sea, new supply bases north of Stavanger were established. Figure A3 in the Appendix 1 Several of the early offshore oil fields were only marginally closer to Norway than the United Kingdom, whose oil fields had been discovered in 1964 and its oil refineries and transport hubs were established before the first Norwegian oil discoveries. For example, the Ekofisk field is 320 km southwest of Stavanger in Norway and 350 km northeast of Teesside in the UK. 5

7 plots the oil industry supply bases from where supplies are delivered to the offshore platforms and oil and gas are landed. These supply bases include Tananger and Dusavik (close to Stavanger), Sotra and Mongstad (close to Bergen), along with Florø, Kristiansund, Sandnessjøen, and Hammerfest. Given the steady increase in new discoveries, the oil shock in Norway was not short lived, but mainly represents a semipermanent income shock, which lasted for more than 40 years until the most recent decline in oil prices in Further, while the initial boom was mainly concentrated in southwestern Norway, today many areas along the western and northern coasts of Norway benefit from nearby oil and gas deposits. Hence, the strong geographic differences in the oil boom are most pronounced in the 1970s. 3 Empirical Strategy Intergenerational mobility is the relationship between outcomes, such as earnings or education in one generation, and the outcomes of the offspring generation. In this section, we first describe our empirical strategy to measure intergenerational mobility. We then detail how we identify the effect of the oil boom on intergenerational mobility. 3.1 Measuring Intergenerational Mobility We employ two different measures of intergenerational mobility: relative mobility and absolute mobility (Chetty, Hendren, Kline, and Saez, 2014). The two measures are related, but provide answers to different questions. Relative mobility measures the average difference in outcomes between children from higher versus lower socioeconomic backgrounds; absolute mobility shows the outcomes of children from a specific (fixed) family background. The most commonly used measure of relative mobility is intergenerational elasticity. This measure computes the percentage change in the income of a son given a 1 percent change in the income of the parents and is estimated by regressing the log earnings of the son s on the log earnings of the father f: log(earnings s i ) = α + β log(earnings f i ) + ɛ i. (1) The slope coefficient β is the intergenerational persistence parameter, with larger values of β indicating a stronger link between parents and sons and thus less mobility. In a society with no intergenerational mobility, we would observe a persistence parameter of β = 1, in one with no relationship between parent son earnings, we would observe a persistence parameter of β = 0. Intergenerational mobility is measured by 1 β, which represents a measure of regression to the mean in percentage terms. Hence, if this elasticity is constant across generations, it is a measure of how many generations it takes for a family living in poverty to reach the average income level. 6

8 However, the intergenerational earnings elasticity is not suited for comparison between subgroups (Mazumder, 2016). In our case, by computing the intergenerational elasticity at the local labor market level, we would compute the regression to the mean within the local labor market region. This does not necessarily allow for meaningful comparisons. For example, suppose that we aim to compare the intergenerational mobility in two local labor market regions A and B. In other words, suppose that the income distribution of the parents generation is the same in the two regions, while the income distribution of children in labor market B is shifted to the right, so that all individuals in B are better off. In this case, it is possible that the regression to the mean in both labor market regions A and B is the same even though the offspring generation in local labor market region B is much richer. Furthermore, the slope coefficient β may not only differ across regions because the correlation in income between generations is different in the two regions, but may also differ if there is a difference in the ratio of the standard deviation of the income distribution of fathers and the standard deviation of the income distribution of sons (Solon, 1999). 2 Estimates of the intergenerational elasticity do not distinguish between these two effects. Hence, the intergenerational earnings elasticity is not a good measure for the comparison between regions when the earnings distributions are different in the different regions and is therefore not useful for our purpose. To compute a measure of relative intergenerational mobility, which allows for a better comparison across local labor markets, we need to standardize the earnings distribution at the national level. A possible solution is to use the rank of individuals in the national income distribution as an outcome, rather than the earnings levels (Mazumder, 2016). Hence, we regress the rank of the son in his earnings distribution on the rank of the father in his earnings distribution: ranki s = α + β rank f i + ɛ i. (2) Assuming that the rank-rank relationship is linear, the estimated parameter β represents the intergenerational persistence in the rank in the earnings distribution. More precisely, β is a measure of the relationship between the position of the sons and the position of the parents in the national earnings distribution of the respective cohorts. A major advantage of measuring intergenerational mobility using rank-rank regression is that the measure is not sensitive to zero incomes and less sensitive to the age when income is measured (Chetty, Hendren, Kline, and Saez, 2014; Nybom and Stuhler, 2016). The intercept α measures the expected rank for sons from fathers at the bottom of the income distribution. 2 This is because the coefficient β can be rewritten as β = Cov(earningss i, earnings f i ) V ar(earnings f i ) = Cov(earningss i, earnings f i ) σ f σ s σ s σ f = Corr(earnings s i, earnings f i ) σs σ f, where σ f is the standard deviation of the fathers earnings distribution and σ s is the standard deviation of the sons earnings distribution. 7

9 As mentioned, measures of absolute mobility denote the outcomes of children from a specific family background. The most commonly used measure of absolute mobility is the transition matrix, which provides the probability of children to be in each quantile of the earnings distribution, given that the parents are in a specific quantile. For example, this measure yields the likelihood of a son growing up in a household in the lowest-earnings quintile to reach the top-earnings quintile in his generation. This measure is particularly relevant when comparing different subgroups with different levels of mobility. In particular, it allows us to investigate specifically whether the differences are driven by changes in bottom-up movements (Chetty, Hendren, Kline, and Saez, 2014). That is, the children of poorer families who are able to move to higher earning quintiles. 3.2 Resource Shock and Intergenerational Mobility The discovery of oil is mostly considered a blessing for Norway and the expansion of oil activities has had far-reaching effects on both Norwegian workers and firms. However, did this resource shock break the economic links between fathers and sons and decrease the intergenerational mobility in local labor markets most affected by the oil discovery? Alternatively, did the oil discovery reinforce earning differences for children in different parts of the income distribution, thereby increasing intergenerational persistence? To determine the effect of the oil boom on intergenerational mobility, we use a similar estimation strategy as Feigenbaum (2015) and regress the rank of the son on the rank of the father along with an interaction term between the rank of the father and a dummy variable indicating whether the son was born in a local labor market affected by the oil boom: ranki s = β 0 + β 1 rank f i + β 2 rank f i Oil llm + X iβ 3 + γ llm + ɛ i, (3) where ranki s is the son s rank in his cohort s income distribution and rankf i is the father s rank in the income distribution of fathers. Oil llm is an indicator of whether the son is born in a local labor market affected by the oil boom. γ llm are local labor market fixed effects. X i is a set of individual characteristics including father s age at childbirth and son s birth cohort fixed effects. We cluster standard errors at the municipality of birth to control for common municipality-level shocks. Hence, β 1 is the persistence parameter in local labor markets little affected by the oil boom. However, the key variable of interest is β 2, which measures the increase or decrease in persistence in local labor markets affected by the oil boom. 3.3 Treated and Control Cohorts To assess the effects of the Norwegian oil boom on intergenerational mobility, we require variation in the importance of the oil sector across local labor markets. Local labor markets are aggregations of municipalities, the lowest administrative level, based on commuting patterns, but they are typically smaller than counties, the medium administrative level (Bhuller, 2009). The 46 local labor markets 8

10 in Norway cover the entire county, including urban and rural areas, and include the area in which people mostly live and work. A local labor market consists on average of nine municipalities with an average population of 68,000 individuals. To measure the local importance of the oil boom, we specify the share of employment in the oil industry and supply industries for the oil sector in each local labor market using the 1980 Census data. To identify the oil supply industries, we follow Brunstad and Dyrstad (1997), who show based on recruiting survey data from 1975, 1977, and 1980, that supply industries such as the manufacturing of metal products, machinery and equipment, and construction should be included in the definition of oil-related industries because these industries are both important suppliers to the oil industry and experienced a large demand shock due to the oil boom in the 1970s. Using threedigit industry codes, we included workers in crude petroleum and natural gas production, petroleum refining, the manufacturing of petroleum and coal products, the manufacturing of machinery, which includes the manufacturing of oil and gas well machinery and tools, the manufacturing of transport equipment, which includes the building of ships and boats, and construction other than building construction, which includes oil well drilling in the definition of oil sector jobs. Figure 4 plots the proportion of workers employed in the oil industry in 1980 in each commuting zone. Generally, the areas with most employment in the oil sector are in southwestern Norway, which is closest to the first oil field discovered in 1969 (see Section 2). We then classify local labor markets into one of three regional groups. In particular, we define local labor markets with less than 7.5 percent of employment in the oil industry as where the oil boom had least impact. 3 Hereafter, we refer to these local labor markets collectively as the low-oil region. Local labor markets with more than 10 percent of employment in the oil industry are where the oil boom had greatest impact and are hereafter collectively referred to as the high-oil region. 4 In our main analysis, we focus on cohorts of sons that entered the labor market during the first decades of oil extraction in Norway. Hence, we look at six cohorts born in the 1950s (birth cohorts ). These cohorts entered the labor market in the 1970s; therefore, they are the first cohorts with the potential to benefit from the expansion of the oil industry in Norway during their whole working life. In addition, these cohorts were largely finished with education (or at least enrolled in either vocational or academic high school education) when oil extraction began. Therefore, we can abstract from the effect of the oil shock on human capital investment. 5 is also supported by Figures 2 and 3, which show that the proportion of individuals born in the 1950s finishing academic or vocational high school was stable and similar in local labor markets both little and heavily affected by the oil boom percent is the median value of employment in the oil industry across the 46 local labor markets. Hence, half of the local labor markets among these local labor markets percent is the 75th percentile of employment in the oil industry across the 46 local labor markets. 5 As existing research shows that men tend to drop out of high school in areas affected by resource booms, this distinction is important (Black, McKinnish, and Sanders, 2005b; Cascio and Narayan, 2015; Morissette, Chan, and Lu, 2015). This 9

11 To argue that β 2 represents the effect of the oil boom on the intergenerational links between fathers and sons, we need to document that intergenerational mobility and the exposure to the oil boom by local labor market were unrelated in those generations before the oil shock. The trends in income per capita in the two regions are very similar and nearly parallel each other from 1950 to 1970 (see Figure 1). Per capita income in the high- and low-oil regions began to diverge only after the first oil discoveries in Moreover, there is much randomness in whether one actually makes an oil discovery, and in particular a large discovery, conditional on oil exploration (Cust and Harding, 2014). 6 Accordingly, although the exact location of the oil and gas deposits and the timing of their discovery is exogenous, the decision where to land the offshore oil and where to locate the headquarters of the Norwegian oil company Statoil was a political decision and therefore nonrandom. Moreover, the increase in oil-relevant employment was larger in areas closer to the oil deposits with preexisting machine and shipping industries that could supply products to the oil industry. Therefore, we turn to two additional cohorts born in the 1930s to show that the exposure to the oil boom does not predict intergenerational mobility for the generation entering the labor market long before the oil journey began in Norway. By examining the generation before the oil boom, we can reveal any geographic variation in intergenerational mobility unrelated to the resource shock. If the intergenerational mobility for sons born in the 1930s was significantly related to the oil shocks, it would suggest that β 2 does not actually reveal the effect of the oil boom on intergenerational mobility, but only existing geographic differences in intergenerational mobility. 4 Data The data we use are compiled from several sources. Our primary data source is the Norwegian Registry Data, a linked administrative dataset that covers the whole population resident in Norway up to These data combine different administrative registers including the central population register, the education register, and the tax and earnings register. 7 In addition, a multigenerational register matches Norwegian children to their parents. These data follow individuals over time in a longitudinal design and provide information about place of birth, place of residence, educational attainment, labor market status, occupations, earnings, a set of demographic variables, as well as information on family background. Information on employment, earnings, and place of residence is collected for each individual every year. In what follows, we briefly summarize the sample definitions and describe the variables and summary statistics for our sample. As discussed in Section 3.3, our analysis focuses on different groups of cohorts and their parents. 6 For example, only 2 percent of exploration wells globally have led to so-called giant oil discoveries and the likelihood is not substantially affected by the extent of exploration (Arezki, Ramey, and Sheng, forthcoming). For instance, when the Johan Sverdrup field was discovered in 2010, currently Norway s second-largest field in terms of reserves remaining, it was within three meters of an exploration well first drilled in See Møen, Salvanes, and Sørensen (2003) for a detailed description of the data. 10

12 Men born from 1952 to 1957, who entered the labor market during the first decades of oil extraction in Norway, constitute the cohorts of primary interest. The sample of their fathers includes individuals born from 1917 to This sample consists of 85,927 son father couples in either lowor high-oil regions for whom we observe lifetime earnings. Note we do not consider the father son couples in the region with percent employment in the oil industry. The total sample in all three regions consists of 107,854 father son couples for whom we observe earnings. The sample of the sons of the 1952 to 1957 cohorts in high- and low-oil regions includes men born from 1968 to 2009 and consists of 116,994 father son couples. In addition, we study Norwegian-born sons born between 1932 and This sample consists of 6,894 son father couples. 9 The central population register contains the municipality of birth, which we use to assign an individual to a local labor market (Chetty, Hendren, Kline, and Saez, 2014). Most cohorts we analyze were born before the first oil discoveries. Hence, assigning oil-boom affectedness by place of birth allows us to abstract from any parental moving decisions, which might be affected by oilrelated employment opportunities. The only cohorts born during the oil boom are the sons of the 1952 to 1957 cohorts. We assign these individuals to a local labor market based on the municipality of birth of their father. In a robustness analysis, we also assign individuals to a local labor market based on their place of residence at age 36. The earnings measure is not top-coded and includes labor earnings expressed in constant 1998 Norwegian Kroner (hence adjusted for inflation), taxable sick benefits, unemployment benefits, parental leave payments, and pensions since As lifetime earnings are less affected by transitory fluctuations than earnings in a single year, we proxy fathers lifetime earnings by averaging their fathers earnings between the ages of 50 and 55 years. Hence, the age of the child when the fathers earnings are measured differs by cohorts. As in Chetty and Hendren (2017), we aim to measure the economic resources of parents while the children are growing up. However, as our yearly earnings data only starts in 1967, the oldest cohorts in our sample are 15 years old when parental earnings are measured. The lifetime earnings of children are defined similarly to fathers earnings. However, we measure children s earnings at ages 36 to Earnings between ages 35 to 40 years should provide us with a reasonable proxy for lifetime earnings (Bhuller, Mogstad, and Salvanes, 2016). By this age, most men have completed their education and have entered the labor market. As we observe earnings until 2014, earnings for the sons of the 1952 to 1957 cohorts are measured at age 30. We observe earnings at age 30 for 46 percent of the sons of the cohorts. The lifetime earnings for those sons born in 1932 and 1933 are also measured at ages 36 to Some 21 percent of men born between 1952 and 1957 have fathers born before 1917 for whom we do not observe earnings during their working life. 9 As discussed below, we are able to link these two cohorts born in the 1930s to their fathers using data from the two years of military service they completed, which is not available for other cohorts born before As pointed out by (Solon, 1999; Haider and Solon, 2006), individuals with high lifetime earnings may have steeper earnings profiles at younger ages. Hence, measuring the earnings of the children when young may understate the intergenerational earnings persistence estimates. Chetty, Hendren, Kline, and Saez (2014) show that such a life-cycle bias is small for rank-rank correlations provided that child earnings are measured after age

13 That is, the earnings measures are from 1968 to 1974 largely a period prior to the oil boom. As we only observe earnings in 1967 and after, the share of sons born in 1932 and 1933 for whom we observe their father s earnings is small. Therefore, we follow Pekkarinen, Salvanes, and Sarvimäki (2017) and construct an alternative measure of father s earnings using records from military conscription data. In Norway, military enlistment was mandatory for all men. Hence, our enlistment data include all males born in 1932 and 1933 and can be linked to the population registry using a personal identification number. In addition, the military recorded information on the occupation and municipality of residence for the fathers of each conscript in both these cohorts. We use the information on fathers occupation and municipality of residence to impute earnings for fathers based on information on average salaries by occupation in 735 Norwegian municipalities from the 1948 tax records. This information allows us to construct imputed earnings for almost 80 percent of the fathers of men born in 1932 and Educational attainment is taken from the educational database provided by Statistics Norway. Since 1974, educational attainment is reported annually by the educational institutions directly to Statistics Norway, thereby minimizing any measurement error. For individuals who completed their education before 1974, we use self-reported information from the 1970 Census, which is considered to be very accurate (see, e.g., Black, Devereux, and Salvanes, 2005). On average, the 1952 to 1957 cohorts have 12.4 years of education while the completed years of education among the father s cohorts are 10.3 years. We observe educational outcomes for 78 percent of the sons of the main cohorts of interest. On average, 44 percent and 31 percent of sons born before 1991 have completed academic and vocational high school, respectively. Men are linked to their spouses using the population registry. The disability pension data are reported by the Social Security Administration. The data include information on the date when disability insurance benefits were awarded and the level of benefits received. An individual is defined as being enrolled in disability insurance if he receives benefits once between 1991 and For the third-generation sons, we also use measures of IQ reported by the military. The IQ score is reported in stanine (Standard Nine) units, a method of standardizing raw scores into a nine-point standard scale that has a discrete approximation to a normal distribution, a mean of five, and a standard deviation of two. We focus our analysis on father-son-couples for three main reasons. First, we can only link male recruits in the 1930s to their fathers. Second, only 1.2 percent of the individuals working in the oil sector in our sample are women. Third, women in the main cohorts of interest ( ) are much less attached to the labor market than men. Table 1 contains summary statistics of the demographic and socioeconomic characteristics for the different types of local labor markets. Altogether, the data we gather provide a unique opportunity to examine how natural resource booms affect intergenerational mobility. 12

14 5 Empirical Results 5.1 Relative Intergenerational Mobility We focus first on the six cohorts entering the labor market at the beginning of the Norwegian oil boom ( ) and start measuring intergenerational mobility using a rank-rank specification (see Chetty, Hendren, Kline, and Saez, 2014). We rank each son based on his earnings aged 36 to 41 relative to others in his birth cohort. Fathers are ranked based on their earnings aged 50 to 55 relative to other men with children born in the same cohort. Figure 5 presents a binned scatterplot of the mean percentile rank of sons versus their fathers percentile rank in the local labor markets, which were either little or heavily affected by the oil boom. These binned scatterplots present the raw earnings data, without controlling for local labor market fixed effects. The conditional expectation of a son s rank given his father s rank is nearly linear in the lowest four quintiles of the fathers earnings distribution. In the top quintile of the fathers earnings distribution, the relationship increases sharply. The relationship between father s and son s rank is less steep in the high-oil region. A less-steep curve implies a weaker link between father s and son s rank and less intergenerational persistence. That is, the sons growing up in the local labor markets that benefitted most from the oil boom experience greater intergenerational mobility. Moreover, the estimated intercept for the high-oil region is larger. That is, the expected rank for sons from fathers at the bottom of the earnings distribution is higher in the high-oil region. This suggests that the sons earnings distribution has shifted to the right. 11 To explore whether the effect of the oil boom on intergenerational mobility is not driven by preexisting local-level differences, we plot the mean percentile rank of sons born in 1932 and 1933 versus their fathers percentile rank in Figure 6. Other than the top and bottom of the fathers earnings distribution, the conditional expectation of a son s rank given his father s rank is nearly linear. Unlike Figure 5, the relationship between father s and son s rank is overlapping and equally steep in the low- and high-oil regions. intergenerational mobility between the two types of regions. Hence, there are no obvious preexisting differences in The graphical results in Figures 5 and 6 are only illustrative and do not control for local labor market fixed effects. Table 2 presents the regression results of Equation 3 and thereby the effect of the oil boom on intergenerational mobility after controlling for local labor market fixed effects, the age of the fathers at childbirth, and the son s year of birth fixed effects. 12 The estimated rank-rank correlation is for cohorts born in 1932 and 1933 (Column i) and for the cohorts born in (Column ii). However, cohorts born in the high-oil region who entered the labor market 11 Section 5.4 provides a detailed discussion of the earnings distributions. 12 Table A1 in the Appendix shows that our main cohorts of interest ( ) are not an especially selected sample, with the estimated intergenerational persistence being similar to other groups of cohorts born in the 1950s and early 1960s. 13

15 after 1970 have an estimated persistence parameter of Hence, in those local labor markets most exposed to the oil boom, intergenerational mobility is significantly higher for the cohorts. Comparing the effect of the oil boom to the overall persistence parameter illustrates the magnitude of the impact: the intergenerational persistence in earnings rank is roughly 14 percent lower in high-oil local labor markets. This is about half of the decrease in intergenerational persistence from the 1930s to the 1950s in Norway and Sweden (Pekkarinen, Salvanes, and Sarvimäki, 2017; Björklund, Jäntti, and Lindquist, 2009). The coefficient of the interaction term is also negative, but not significant for the cohorts born in 1932 and Hence, intergenerational mobility was slightly different prior to the oil boom in the two regions. However, there is no significant preexisting difference in intergenerational mobility between the low- and high-oil regions. These findings are in line with the patterns plotted in Figures 5 and 6. Moreover, the main result that a large economic boom increases intergenerational mobility in the areas most affected corresponds well with Feigenbaum (2015) who shows that the Great Depression lowered intergenerational mobility in the US for sons growing up in cities hit hardest by economic downturn. 5.2 Absolute Intergenerational Mobility Relative intergenerational mobility may be driven by both bottom-up movements and by poorer outcomes for the rich (Chetty, Hendren, Kline, and Saez, 2014). Therefore, measuring absolute intergenerational mobility is valuable and will help us understand whether individuals growing up in poor or moderately well-off families benefited most from the natural resource shock. Table 3 presents the intergenerational transition matrices, which show the probability of sons to be in each quintile of the earnings distribution, given that their father is in a specific quintile of the father s earnings distribution in low- or high-oil regions. In the low-oil region, the likelihood of sons growing up in a household in the lowest earnings quintile remaining so is 29 percent and 12 percent attain the top earnings quintile. Conversely, the probability that the sons of the poorest 20 percent of families in the high-oil region remaining in the lowest earnings quintile is 23 percent and that they reach the top earnings quintile is 17 percent. The differences in absolute mobility across the regions are significant at the 1 percent level. Hence, individuals born to poor families in the region that benefitted early from the oil boom are significantly less likely to remain poor and more likely to move all the way to the top earners of their cohort. To ensure these findings are not driven by preexisting local-level differences, Table 4 presents the intergenerational transition matrices for the cohorts born in 1932 and Corresponding with the findings from the rank-rank regression (see Section 5.1), the differences between the lowand high-oil labor markets are less pronounced. The percentage of bottom-to-top (the percentage of sons with fathers in the bottom 20 percent of the earnings distribution who attained the top 20 percent of their earnings distribution) is 12 percent in the high-oil region and 10 percent in 14

16 the low-oil region. As for the relative mobility measure, there is somewhat more upward mobility prior to the oil shock in the oil-affected region, but the regional differences increase substantially following the resource shock. Moreover, the differences in absolute mobility across regions are not significant at the 5 percent level for the birth cohorts. Notably, the oil boom makes it less likely that sons remain in the lower half of the earnings distribution. This holds for sons from all socioeconomic backgrounds except those from the top 20 percent of families. Hence, the oil boom increased bottom-up mobility, but not so much for individuals growing up in the richest families. The question remains whether these upward moving individuals are employed directly in the oil sector. In the oil boom-affected region, 29.7 percent of sons who reached the top earnings quintile irrespective of the father s earnings quintile were working in the oil industry. The same figure is only 9.8 percent in low-oil labor markets. Overall, the percentage of workers employed in the oil sector is about 15 percent in the high-oil region and 6 percent in the low-oil region. Hence, in all labor markets, upward movers display a greater likelihood of working in oil-related industries and more than proportionally so in the high-oil region. 5.3 Educational Attainment and Returns to Education From the perspective of the 1952 to 1957 cohorts, the oil shock can be considered a fortunate event, which increased earnings and intergenerational mobility in the areas most affected. However, the question remains how this increase in earnings and intergenerational mobility is linked to human capital. There are two possible channels. First, the oil boom may have altered the investments in human capital accumulation depending on parental background. Alternatively, the increase in earnings and intergenerational mobility may be caused by changes in the returns to human capital endowments. First, we provide some descriptive evidence documenting the relationship between family background, as proxied by fathers earnings, and the son s accumulated human capital. Figure 7 presents a binned scatterplot of the probability that sons born between 1952 and 1957 attained an academic high school education (or a higher education) by their fathers percentile rank. We find that in both types of labor markets, the sons of richer fathers are more likely to get an academic high school education. The differences between the low- and high-oil labor market regions are small. Hence, the education gap between sons growing up in poor or rich households is not so much affected by the oil boom as for the cohorts born in the 1950s. This is not surprising given the fact that these cohorts mainly completed their education prior to the start of the oil boom and that average education levels are very similar in the high- and low-oil regions (see Figures 2 and 3). Figure 8 presents a binned scatterplot of the probability that sons born in 1932 and 1933 have an academic high school education versus their fathers percentile rank by type of labor market. Also for these cohorts, the differences between the low- and high-oil labor market regions in the probability of having an academic high school degree conditional on father s earnings percentile is small. 15

17 As the graphical results in Figures 7 and 8 do not control for local labor market fixed effects, we use a similar specification as described in Equation 3 to examine whether the earnings rank of the father is less correlated with his son s probability to complete academic high school in the highoil region. Therefore, we regress an indicator variable whether the son completed academic high school on the father s rank in the earnings distribution and include an interaction term with the father s rank and another for being born in the high-oil region. We include controls for local fixed effects, for the age of the fathers at childbirth, and the son s year of birth fixed effects. Figures 7 and 8 suggest that the relationship between the rank of fathers and the probability of sons finishing academic high school is not linear. For this reason, we augment this specification with a quadratic function of the fathers rank and the interaction between the fathers rank and the high-oil region dummy (see Table 5). The coefficient for the quadratic function of the fathers rank is positive and significant in all specifications, implying that the relationship between education and the earnings of fathers is flatter at the bottom of the earnings distribution and steeper for richer fathers. The parameter for the interaction term between the rank of fathers (and for the quadratic form of this rank) and the dummy for living in areas highly affected by the oil boom is not significant for cohorts born from 1932 to 1933 and from 1952 to Hence, there are no regional differences in human capital investment conditional on fathers earnings. These results provide empirical evidence that the oil boom did not lead to changes in human capital investment conditional on fathers earnings, which could explain the change in the intergenerational mobility observed for the cohorts born in the 1950s. Second, returns to accumulated human capital may have changed as a consequence of the oil shock. Table 6 presents the results from a regression of log average earnings at ages 36 to 41 on a dummy variable for completing academic high school or college before age 36, an interaction of the academic education indicator and dummy variables for being born in the high-oil region, cohorts, and local labor market fixed effects. Average earnings are 35 percent (cohorts ) or 45 percent (cohorts ) higher for individuals with an academic education compared to those with a vocational high school degree or without an academic high school degree. However, for cohorts entering the labor market at the beginning of the oil boom, there are significant geographical differences in the returns to academic education. For men born in the high-oil region in , the returns to academic education average 37 percent (8 percentage points) lower. Notably, the interaction term between the academic education indicator and the high-oil region indicator is not significant for the 1932 and 1933 birth cohorts. As exploitation of natural resources mostly creates jobs for skilled and semiskilled craftsmen and low-skilled workers, the lower returns to academic education are not surprising. A larger demand for vocationally trained workers increases their price and thus reduces the returns to higher education. Different types of natural resource shocks to local labor markets in other countries have accounted for similar changes in the returns to education (see, e.g., Cascio and Narayan, 2015; Emery, Ferrer, 16

18 and Green, 2012). As discussed by Becker and Tomes (1976) and shown in Figures 7 and 8, individuals from less affluent backgrounds have a lower probability of finishing academic education. Hence, the sons of poorer fathers benefiting more than proportionally from the increasing returns to vocational training in the high-oil region leads to greater upward mobility in that region. The hypothesis that changes in returns to human capital is the main driver of earnings mobility has been pointed out by Aaronson and Mazumder (2008), who show that the increase in the intergenerational elasticity observed in the US during the 1980s and 1990s corresponds with the period of increased returns to college. In a cross-country comparison, Corak (2013) finds that countries with the largest college wage premium are less mobile across generations. In contrast, our finding that changes in the returns to education are an important driver of the increase in intergenerational mobility somewhat differ from Feigenbaum (2015), who shows that the Great Depression affected intergenerational mobility through returns to the endowed family capital. 5.4 Cross Sectional Inequality and Intergenerational Mobility As discussed in Section 5.1, the increased intergenerational mobility for cohorts born between 1952 and 1957 in the high-oil region implies that the earnings distribution in this region shifted to the right when oil production in Norway began. Figure 9 plots the estimated earnings distribution by region for the cohorts born in 1932 and 1933 along with the cohorts born from 1952 to The earnings distributions in the high- and low-oil regions mainly overlap for the 1930 cohort and the two distributions are not significantly different from each other. However, the earnings distribution for the cohorts born in the high-oil region who entered the labor market at the beginning of the oil boom shifts to the right in the lower and middle part of the earnings distribution. Only for high earnings do the two distributions overlap. Both distributions are significantly different. The dispersion indexes in Table 7 confirm these findings. For the cohorts born in 1932 and 1933, there are no substantial differences in earnings inequality between the two regions. In addition, for cohorts born between 1952 and 1957, the ratio of the 90th and 50th percentiles measuring inequality in the upper part of the earnings distribution does not differ by region. However, the ratio of the 10th and 50th percentiles measuring inequality in the lower part of the earnings distribution is higher in the high-oil region. This decreased dispersion in the left part of the earnings distribution translates to a lower Gini index for the earnings distribution for workers in the high-oil region (0.27 compared to 0.30 in the low-oil region). Hence, in areas where low earnings were pushed upwards and inequality decreased, intergenerational mobility increased. This shift in the lower part of the earnings distribution in the high-oil region reflects the findings from both the analysis of relative (Section 5.1) and absolute mobility (Section 5.2). First, it is consistent with the larger estimated intercept in the rank-rank regressions for the high-oil region implying that the expected rank for sons from fathers at the bottom of the earnings distribution is higher in that region. Second, it reflects the substantial increase in absolute bottom-to-top mobility in the 17

19 high-oil region. As described by Hassler, Mora, and Zeira (2007), changes in inequality are not necessarily negatively associated with changes in intergenerational mobility. In particular, an increase in inequality may increase the incentive to undertake more education and thereby account for greater mobility. However, in the case of a resource shock, the increased returns to less educational attainment raises the earnings for low-skilled workers and may thereby lead to a decrease in inequality and an increase in mobility. The decreased returns to academic education imply that the earnings distribution in the high-oil region for vocationally educated individuals shifted especially to the right when oil production began. Figure 10 shows that almost the entire earnings distribution for vocationally trained individuals in the high-oil region shifted to the right. Only at the very top does the distribution overlap with the earnings distribution for vocationally trained individuals in the low-oil region. Similarly, the earnings distribution of academically educated workers slightly shifted to the right, implying that highly educated workers also earn more in areas most affected by the oil boom. Moreover, among vocationally trained individuals, the differences in the earnings distributions by region are greatest for individuals working in the oil industry (see Figure 11). However, the earnings distribution for vocationally trained individuals working in non-oil industries also shifted to the right in the high-oil region. This indicates the presence of earnings spillovers from oil to nonoil sectors. That is, individuals working in industries supplying goods and services to the oil industry workers (e.g., hairdressers, restaurant employees, cleaning services, car dealers, etc.) also benefit from the higher earnings in the resource sector. Nonetheless, among individuals with an academic education, only those working in the oil sector experienced a shift in the earnings distribution. This might be explained by the fact that many academically trained individuals in the nonoil sector are public sector workers (e.g., teachers, medical doctors, etc.) paid according to a nationwide pay scale. 5.5 Multigenerational Persistence We presented empirical evidence that a natural resource shock increases intergenerational mobility for cohorts directly affected early in their working life. However, the question remains whether this natural resource shock is strong enough to also affect the children of the affected cohorts. Lindahl, Palme, Massih, and Sjögren (2015) and Clark (2014) discuss that estimates of intergenerational persistence obtained from data on two generations severely underestimate long-run intergenerational persistence. Moreover, Nybom and Stuhler (2013) argue that periods of structural change may reduce the transmission of social status in the directly affected generation. However, this decrease in intergenerational persistence is not necessarily still in place for the next generation because family ties may again tighten and the society could enter a new steady state with lower intergenerational mobility. Figures 13 and 14 present some descriptive evidence of the intergenerational mobility across 18

20 multiple generations. In particular, the figures present binned scatterplots of the mean percentile rank of third-generation sons versus their grandfathers and fathers percentile rank in high- and low-oil regions. Sons are ranked based on their earnings at age 30 relative to other men with fathers born in the same cohort. The conditional expectation of a son s rank given his grandfather s rank is nearly linear in the lowest four quintiles of the grandfathers earnings distribution. In the top quintile of the grandfathers earnings distribution, the relationship decreases sharply. In the highoil region, the relationship between grandfather s and son s rank is nearly flat, implying that there is no link between the grandfather s and son s earnings rank. That is, we find no intergenerational persistence between grandfathers and sons in local labor markets affected by the oil boom. In the low-oil region, there is a positive relationship between the grandfather s and son s rank. Hence, sons with higher-earning grandfathers in the low-oil region are on average earning higher earnings at the age of 30 than sons with lower-earning grandfathers. In addition, the estimated intercept for the high-oil region is larger, indicating that the expected rank for sons from grandfathers at the bottom of the earnings distribution is higher. Figure 14 plots the mean percentile rank of third-generation sons versus their fathers percentile rank. Besides the top and bottom part of the fathers earnings distribution, the conditional expectation of a son s rank given his father s rank is nearly linear. Unlike Figure 13, the relationship between father s and son s rank has a positive gradient in both types of region. However, the relationship between father s and son s rank is less steep in the high-oil region, implying a weaker link between the father s and son s rank and less intergenerational persistence. This indicates that the higher intergenerational mobility caused by the oil boom in the high-oil region persists over multiple generations. As Figures 13 and 14 are only illustrative and do not control for local labor market fixed effects, Table 8 presents the regression results of Equation 3. The regressions control for local labor market fixed effects, for the age of the grandfathers at childbirth, and the father s year of birth fixed effects. The estimated rank-rank correlation is for the grandfather son comparison (Column i) and for the father son comparison (Column ii). In the high-oil region, the estimated persistence parameter for the father son comparison averages 32 percent (5.4 percentage points) lower. For the grandfather-son comparison, we identify zero persistence in the high-oil region. Hence, the oil boom has broken the earnings link between grandfathers and sons in the oil boom-affected region. These results correspond with the findings in the graphical analysis and show that the oil boom affected intergenerational mobility across multiple generations. Next, we analyze the third-generation sons absolute mobility. Figures 15 and 16 plot the mean rank of third-generation sons conditional on their grandfathers and fathers rank in the high- and low-oil regions. 13 The average ranks for sons whose father grew up in a poor family and remained in a low earnings percentile are higher in the high-oil region. Sons whose fathers grew up in relatively poor families (but not in the lowest percentile) and themselves moved to the top of the 13 The transition matrixes corresponding to Figures 15 and 16 are presented in Table A4 in the Appendix. 19

21 earnings distribution achieve a substantially higher rank in the high-oil region. However, there are no regional differences in average ranks for sons whose father grew up in a very rich family and remained in the top earnings percentile. Hence, the sons of bottom-to-top movers benefit most in terms of early career earnings from the natural resource shock. 14 To summarize, we find that persistence decreases over generations in both high- and low-oil regions and that intergenerational mobility is significantly higher in the high-oil region. This differs from the findings of Nybom and Stuhler (2013), who show that an educational reform in Sweden increased intergenerational mobility in income and education from parents to their offspring in the directly affected generation, but increased intergenerational persistence in the next generation. A possible explanation for the difference in results is the very long duration of the Norwegian oil boom. The boom not only affected the cohorts born in the 1950s but also their offspring. Hence, the structural change that reduces the transmission of social status lasted for more than 40 years and society may not yet have entered a new steady state with less intergenerational mobility, at least in the high-oil region. 15 However, the oil price dramatically fell in the second half of Just from 2014 to 2015, direct and indirect employment in the oil sector fell by 35,000 workers, corresponding to an 18 percent decrease in the workforce in the oil sector and a large share of the laid-off workers were skilled and semiskilled craftsmen. Depending on the development of the oil price and productivity in the oil sector, the impacts on the third-generation s earnings could differ in the future. To obtain a better understanding of how vulnerable the sons of bottom-to-top movers are with respect to job prospects in the oil sector, we plot the probability of third-generation sons to have an academic education conditional on their grandfathers and fathers rank in the high- and lowoil regions in Figures 17 and On average, sons whose father grew up in a poor family and remained in a low earnings percentile have similar education levels in the high- and low-oil regions. There are also small regional differences in the average education level for sons whose father grew up in a very rich family and remained in the top earnings percentile. However, sons whose fathers grew up in relatively poor families and themselves moved to the top of the earnings distribution have on average a significantly higher education in the low-oil region. That is, bottom-to-top movers in the high-oil region invest less in the education of their offspring than do bottom-to-top movers in the low-oil region. Note that for the same comparison in Figures 15 and 16, sons in the high-oil region earn substantially more. That is, while the sons of bottom-to-top mover fathers 14 As we can only observe earnings until 2014, we measure the earnings of the third generation at age 30. As discussed earlier, earnings between ages 36 and 40 would be a better measure of lifetime earnings as individuals with college education have much less experience at age 30 than vocationally trained individuals, but show a steeper earnings profiles in their early 30s. Figures A5-A8 and Tables A8-A9 in the Appendix show the results for Section 5.5 where the earnings for the third-generation results are adjusted for experience. In particular, we predict earnings at age 30 based on a second-order polynomial function of experience. The results scarcely deviate from those presented above. 15 Note that our latest earnings measure for the third-generation sons is from 2014 (see Section 4). 16 The transition matrix corresponding to Figures 17 and 18 are presented in Table A10 in the Appendix. 20

22 benefit most in terms of early career earnings, their education level is lower. Despite the lower educational investment, these individuals achieve high earnings through high returns to vocational training. Should the returns to vocational training fall further because of yet lower oil prices and mass layoffs in the oil sector, the sons of the bottom-to-top movers are more likely to suffer. There are different possible reasons for the lower investment in education of their offspring of the bottom-to-top movers in the high-oil region. First, bottom-to-top movers in the low-oil region are better educated (see Table A6 in the Appendix). That is, bottom-to-top movers in the low-oil region are more likely to do much better than their fathers did because of their education. Hence, their experience might motivate them towards the education of their sons or their sons might have their fathers as role models and more often aspire to enter academic education. Second, family formation could be important. Figure A4 in the Appendix shows that second-generation men in the high-oil region are on average slightly younger when their have their first child and Table A7 in the Appendix shows that the spouses of bottom-to-top movers in the high-oil region are less likely to have an academic education. Thus, the mothers of the bottom-to-top movers sons in high-oil region may also be inclined towards encouraging an academic education. Third, the difference in education could also reflect differences in ability. Figures A9 and A10 plot the third-generation sons IQ conditional on their grandfathers and fathers percentile rank in the high- and low-oil regions. Men whose grandfathers were poor score substantially lower in the IQ tests at age 19 in the high-oil region compared to the low-oil region. Moreover, having a father in the highest percentile ranks is less predictive of IQ score in the high-oil region independent of the grandfather s percentile rank. As IQ scores not only reflect genetic traits but also early parental investments, this suggests that the sons of bottom-to-top movers might have lower genetic and early-acquired ability. As noted earlier, most individuals working in the oil industries or industries supplying to the oil industry are men. The education and career decisions of men and women conditional on their family background might therefore be substantially different. Hence, we also analyze the intergenerational earnings mobility and educational outcomes for third-generation daughters. For children of the cohorts, the likelihood of being in the labor force is about equally high for sons and for daughters. Table A11 in the Appendix presents the estimated rank-rank correlations between grandfathers and daughters as well as fathers and daughters. For the father daughter comparison (Column ii), the results are not significantly different to the father son comparison. The estimated rank-rank correlation is (0.166 for the father son comparison in Table 8) and the estimated persistence parameter is on average lower in the high-oil region. For the grandfather daughter comparison, we find greater intergenerational persistence than for the grandfather son comparison, but it does not differ by oil region. Hence, the oil boom did not break the earnings link between grandfathers and daughters in oil boom-affected regions. Figures A13 and A14 in the Appendix plot the average earnings percentile rank of daughters condi- 21

23 tional on the earnings percentile rank of their father and their grandfather. Compared to Figures 15 and 16, the grandfather s earnings percentile particularly when grandfathers are in the highest percentile ranks seems to be a more-important determinant of the daughter s earnings percentile than for sons in both high- and low-oil regions. Moreover, the average earnings percentile of daughters of bottom-to-top movers is not significantly different in the two types of region (see Table A12 in the Appendix). However, the pattern for daughters education conditional on the earnings percentile rank of fathers and grandfathers is similar to the pattern for sons (see Figures A15 anda16 and Table A13 in the Appendix), suggesting the daughters of bottom-to-top movers acquire less academic education in the high-oil region. Hence, this result confirms that families of bottomto-top movers in the high-oil region are less encouraging towards education than the families of bottom-to-top movers in the low-oil region. 6 Robustness Analysis 6.1 Internal Migration As mentioned in Section 4, we assign each individual to a local labor market based on their municipality of birth to avoid the possibility that selection into migration because of the oil shock biases our results. The extant literature demonstrates that local economic booms can affect migration. For example, Arthi, Beach, and Hanlon (2017) find that the relationship between business cycles and mortality is highly sensitive to migration and Dinkelman (2011) shows that electrification in rural South Africa affected migration behavior and changed the skill composition of the workforce in newly electrified areas. Even though we assign the local labor market based on the municipality of birth, our results might still be biased through selective migration. For example, if a large share of men from a poor family background move from a low-oil region to a high-oil region where they earn more than in their region of birth, and thereby move up in the national earnings distribution, the regional differences in intergenerational earnings persistence would be underestimated. We would also underestimate the regional differences in the intergenerational earnings persistence if the sons of richer fathers migrate out of a high-oil region and remain in the top part of the national earnings distribution. However, we would overestimate the regional differences in intergenerational earnings persistence if men from a poor family background migrate from the high-oil region to the low-oil region and earn more. To analyze how selective migration could affect our estimates of intergenerational mobility, we proceed in three steps. First, we document the number and characteristics of migrants moving from the low-oil region to the high-oil region and vice versa. Second, we estimate our main analysis based on the sample of sons remaining in the region of birth during their working life. Third, we assign each individual to a local labor market based on their place of residence when they are aged 36 and re-estimate our main analysis. 22

24 An individual is defined as a mover if he is registered as resident in a different region as the region of birth for at least 1 year between ages 18 and 41. Table 9 details descriptive statistics for movers and stayers by type of region. Note that the proportions do not sum to one because we do not consider individuals moving to or from local labor markets where the oil-related employment share lies between 7.5 and 10 percent (see Section 3.3). Overall, men born in the 1950s are more likely to move than are men born in the 1930s. The number of stayers is substantially higher in the low-oil region both for the cohorts born between 1952 and 1957 and those born in 1932 and Men born in the 1950s in the high-oil region are substantially more likely to move to the low-oil region than men born in the low-oil region are to move to the high-oil region. Hence, we do not observe a large stream of migrants moving towards the region where the oil sector was booming. 17 In addition, movers are more likely to be from a richer and better-educated family background and they themselves are better educated. Table 10 presents the regression results of Equation 3 for three different samples. Column (i) is our baseline specification where an individual is assigned to a local labor market based on the municipality of birth. In Column (ii), movers are excluded and in Column (iii) we assign an individual to a local labor market based on the municipality of residence at age 36. In the regressions, we control for local labor market fixed effects, for the age of the fathers at childbirth, and the son s year of birth fixed effects. The estimated rank-rank correlations in all three columns are similar, but lowest when we assign individuals to their place of residence instead of their place of birth. The estimated regional difference in the rank-rank correlation is also smallest when we assign individuals to their place of residence. However, the difference between the coefficient of the interaction term β 2 is not significant when comparing Columns (i) and (iii) and the estimated intergenerational persistence parameters in the high-oil region shifts only from to when reassigning the local labor market. Excluding all migrants leads to even greater differences between the estimated intergenerational persistence parameters in the high- and low-oil regions (see Column ii). Tables A2 and A3 in the Appendix present the transition matrices for the sample excluding all movers and where the local labor market is assigned based on the place of residence. When excluding movers, the likelihood of sons growing up in a household in the lowest earnings quintile to remain in the lowest earnings quintile is similar to the estimates in Table 3 for both types of region. However, the likelihood of sons growing up in a household in the lowest earnings quintile to reach the top earnings quintile is lower than in our baseline sample in the high-oil region. Moreover, the likelihood of sons growing up in a household in the highest earnings quintile to remain in the top earnings quintile is lower than in our baseline sample in the high-oil region. Comparing the results in Table A3 (assignment based on residency) with our baseline estimates in Table 3 (assignment based on place of birth), 17 Bartik (2017) shows that for the case of fracking in the US, workers do not respond to a local resource boom as predicted by a model with no costs of moving. He concludes that the type of workers who would benefit from a labor demand shock given a resource boom have positive costs of moving. 23

25 the differences are small. Hence, we find some indication that the actual difference between the intergenerational persistence parameters in the high- and low-oil regions might be either smaller or larger, but the biases are not large and the significant differences between the two types of oil regions remain. These findings differ from Feigenbaum (2015), who finds that migration is a key mechanism for his results that the sons of richer fathers migrated to locations that had suffered less-severe effects from the Great Depression. However, the difference in the importance of the drivers of intergenerational transmission is not surprising because Feigenbaum (2015) analyzes a different type of economic shock and a bust instead of a boom period. 6.2 Excluding the Oslo Labor Market Region Compared to other local labor markets, Oslo is substantially larger in terms of inhabitants and economic power. To ensure that Oslo is not driving our results, Column (iv) in Table 10 presents the regression results of Equation 3 after excluding all men born in Oslo. Both the estimated intergenerational earnings persistence and the coefficient of the interaction term are slightly smaller when excluding Oslo. However, the regional difference in intergenerational persistence is not significantly different from the baseline results Column (i) in Table Adverse Health Effects Moving up the income ladder as a less-skilled worker might come at a health risk. In particular, oil industry jobs for less-skilled men are potentially riskier and there is also a risk of adverse health effects from working in shifts offshore. Hence, bottom-to-top movers in the high-oil region might be more likely to receive disability insurance payments. Table 11 shows the probability of receiving disability insurance payments at least once between 1991 and 2008 conditional on both the son s and the father s earnings quintile. On average, the percentage of individuals ever receiving disability insurance payments is higher in the high-oil region for sons in the first and second quintile of their earnings distribution. However, sons in the two highest quintiles of their earnings distribution in the high-oil region are less likely to receive disability insurance payments. This also holds for bottomto-top movers in the high-oil region who are significantly less likely to ever enroll in disability insurance compared to the bottom-to-top movers in the low-oil region. Hence, we do not find any evidence that bottom-to-top movers in the high-oil region gain their earnings predominantly through jobs exposing them to health risks An alternative measure of occupational health risk would be fatal accidents or death due to occupational health hazards (e.g., exposure to certain chemicals). However, the number of fatal accidents in the oil industry was not much larger than for other industries with predominantly skilled and semiskilled craftsmen (e.g., construction, transport, machinery, etc.). The individuals in the cohorts were still too young in 2014 to observe a lower life expectancy due to occupational health hazards. 24

26 6.4 Intergenerational Mobility within Regions As discussed in Section 3.1, we standardize the earnings distribution at the national level to allow for comparisons across local labor markets (see Mazumder, 2016). However, in Section 5.4 we show that the increase in intergenerational mobility in the high-oil region is largely caused by a shift in the lower part of the earnings distribution in that region. Hence, the question remains how the oil boom affected intergenerational mobility within each region over time. To measure the intergenerational mobility within each type of region over time, we rank individuals within either the high- or low-oil regions. In the high-oil region, the conditional expectation of a son s rank given his father s rank is for the cohorts born in the 1930s and for the cohorts born in the 1950s. That is, within the high-oil region, intergenerational mobility was not much affected by the oil boom. For the sons of the cohorts born in the 1950s in the high-oil region, the conditional expectation of a son s rank given his father s rank is Hence, intergenerational mobility within the high-oil region increased for the third generation. For all three comparisons in the low-oil region, the estimated persistence parameters are higher. Intergenerational mobility is also increasing over time in the low-oil region, but the increase is substantially slower than in the high-oil region. Hence, the regional differences in intergenerational mobility caused by the oil boom are mainly driven by shifts in the lower part of the earnings distribution in the high-oil region for the cohorts. For the sons of the cohorts, we observe an increase in within-region mobility. 7 Conclusion In this paper, we analyzed the impact of a large economic shock on intergenerational mobility. In particular, we studied how the Norwegian oil boom starting in the 1970s affected the transmission of economic status between sons who entered the labor market around the start of the oil boom and their fathers. Our results indicate that the oil boom increased intergenerational earnings mobility: sons born in local labor markets that benefitted most from the oil boom in the 1970s experienced more intergenerational earnings mobility than sons born in the region that benefitted least (in terms of employment). These effects are not driven by preexisting location-specific differences in intergenerational earnings mobility. The transmission of economic status for cohorts who did not benefit from the oil boom early in their working career does not differ significantly between the high- and low-oil regions. The new economic opportunities were also beneficial for men across all socioeconomic backgrounds, but mostly increased bottom-up mobility. We also find empirical evidence that geographic differences in human capital investment are not the mechanism underpinning our findings. However, the higher bottom-up mobility might be explained by the significantly lower returns to academic education in the regions most affected by the oil boom. Moreover, our results indicate that the oil boom shifts the lower part of the earnings distribution in the oil-affected region towards the right and that this shift is mostly driven by 25

27 vocationally trained workers in the oil sector. The results are not sensitive to the exclusion of the Oslo labor market, to adverse health effects, or to selective migration. This last finding differs from Feigenbaum (2015), who finds that migration is a key mechanism because the sons of richer fathers migrated to locations that had suffered less-severe effects from the Great Depression. However, the difference in the importance of the drivers of intergenerational transmission is not surprising because Feigenbaum (2015) analyzes a different type of economic shock and a bust instead of a boom period. Analyzing intergenerational mobility across three generations, we find that the main results persist for a third generation. Although the estimated intergenerational persistence decreases over generations in both the high- and low-oil regions, intergenerational mobility is still significantly higher in the high-oil region in both the grandfather son and father son comparison. Hence, the oil boom persistently broke the earnings link between generations. As the oil boom lasted for more than 40 years and potentially also affected the third-generation s earnings, it is not surprising that our results differ from Nybom and Stuhler (2013), who show that an educational reform in Sweden increased intergenerational mobility in earnings and education from parents to their offspring in the directly affected generation, but increased intergenerational persistence in the next generation. While the sons of bottom-to-top movers in the oil-affected region on average have the highest early career earnings, their level of education is significantly lower, and they are therefore more vulnerable to negative oil shocks and decreasing returns to vocational education. As the oil price dropped dramatically in the second half of 2014 and the workforce particularly the less skilled in the Norwegian oil sector decreased substantially, the third-generation s earnings, and thereby also intergenerational earnings persistence, could change during the next decade. Therefore, future research should seek to analyze how shorter-lived economic shocks affect intergenerational mobility across multiple generations. 26

28 References Aaronson, D., and B. Mazumder (2008): Intergenerational Economic Mobility in the United States, 1940 to 2000, Journal of Human Resources, 43(1), Allcott, H., and D. Keniston (forthcoming): Dutch Disease or Agglomeration? The Local Economic Effects of Natural Resource Booms in Modern America, Review of Economic Studies. Arezki, R., V. A. Ramey, and L. Sheng (forthcoming): News Shocks in Open Economies: Evidence from Giant Oil Discoveries, Quarterly Journal of Economics. Arthi, V., B. Beach, and W. W. Hanlon (2017): Estimating the Recession-Mortality Relationship when Migration Matters, Working Paper 26507, National Bureau of Economic Research. Bartik, A. (2017): Worker Adjustment to Changes in Labor Demand: Evidence from Longitudinal Census Data, Working paper. Basso, G. (2016): Local Labor Markets Adjustments to Oil Booms and Busts, Working paper. Becker, G. S., and N. Tomes (1976): Child Endowments and the Quantity and Quality of Children, Journal of Political Economy, 84(4, Part 2), S143 S162. Bhuller, M. (2009): Inndeling av Norge i arbeidsmarkedsregioner, Notater 2009/24, Statistics Norway. Bhuller, M., M. Mogstad, and K. G. Salvanes (2017): Life Cycle Earnings, Education Premiums and Internal Rates of Return, Journal of Labor Economics, 35(4). Björklund, A., M. Jäntti, and M. J. Lindquist (2009): Family Background and Income During the Rise of the Welfare State: Brother Correlations in Income for Swedish Men Born , Journal of Public Economics, 93(5), Black, D. A., T. McKinnish, and S. Sanders (2005a): The Economic Impact Of The Coal Boom And Bust, The Economic Journal, 115(503), Black, D. A., T. G. McKinnish, and S. G. Sanders (2005b): Tight Labor Markets and the Demand for Education: Evidence from the Coal Boom and Bust, Industrial and Labor Relations Review, 59(1), Black, S. E., P. J. Devereux, and K. G. Salvanes (2005): The More the Merrier? The Effect of Family Size and Birth Order on Children s Education, The Quarterly Journal of Economics, 120(2),

29 Brunstad, R. J., and J. M. Dyrstad (1997): Booming Sector and Wage Effects: An Empirical Analysis on Norwegian Data, Oxford Economic Papers, 49(1), Cascio, E. U., and A. Narayan (2015): Who Needs a Fracking Education? The Educational Response to Low-Skill Biased Technological Change, Working Paper 21359, National Bureau of Economic Research. Chetty, R., and N. Hendren (2017): The Impacts of Neighborhoods on Intergenerational Mobility I: Childhood Exposure Effects, Working Paper 23001, National Bureau of Economic Research. Chetty, R., N. Hendren, P. Kline, and E. Saez (2014): Where is the land of Opportunity? The Geography of Intergenerational Mobility in the United States, The Quarterly Journal of Economics, 129(4). Chetty, R., N. Hendren, P. Kline, E. Saez, and N. Turner (2014): Is the United States still a Land of Opportunity? Recent Trends in Intergenerational Mobility, The American Economic Review, 104(5), Clark, G. (2014): The Son Also Rises: Surnames and the History of Social Mobility. Princeton University Press, Princeton. Cooper, B., and T. Gaskell (1976): Adventure of North Sea Oil, Hardcover. William Heinemann Ltd, London, 1 edn. Corak, M. (2013): Income Inequality, Equality of Opportunity, and Intergenerational Mobility, The Journal of Economic Perspectives, 27(3), Corak, M., M. J. Lindquist, and B. Mazumder (2014): A comparison of Upward and Downward Intergenerational Mobility in Canada, Sweden and the United States, Labour Economics, 30, Cust, J., and T. Harding (2014): Institutions and the Location of Oil Exploration, Working Paper 127, University of Oxford. Dinkelman, T. (2011): The Effects of Rural Electrification on Employment: New Evidence from South Africa, The American Economic Review, 101(7), Ekeland, A. (2015): Sysselsatte i petroleumsnæringene og relaterte næringer 2014, Reports 2015/48, Statistics Norway. Emery, H., A. Ferrer, and D. Green (2012): Long Term Consequences of Natural Resource Booms for Human Capital Accumulation, Industrial and Labor Relations Review, 65(3),

30 Feigenbaum, J. (2015): Intergenerational Mobility during the Great Depression, Working paper. Finansdepartementet (1974): Petroleumsvirksomhetens plass i det norske samfunn. Norwegian Government, Oslo. Haider, S., and G. Solon (2006): Life-Cycle Variation in the Association between Current and Lifetime Earnings, American Economic Review, 96(4), Hassler, J., J. V. R. Mora, and J. Zeira (2007): Economic Growth, 12(3), Inequality and Mobility, Journal of Helle, E. (1984): Norges Olje de frste 20 årene. Tiden, Oslo. Lerøen, B. V. (1990): Fra Groningen til Troll: Norske Shell - 25 år på norsk sokkel. Norske Shell, Stavanger. Lindahl, M., M. Palme, S. S. Massih, and A. Sjögren (2015): Long-Term Intergenerational Persistence of Human Capital: An Empirical Analysis of Four Generations, Journal of Human Resources, 50(1), Løken, K. V. (2010): Family Income and Children s Education: Using the Norwegian Oil Boom as a Natural Experiment, Labour Economics, 17(1), Mazumder, B. (2016): Estimating the Intergenerational Elasticity and Rank Association in the United States: Overcoming the Current Limitations of Tax Data, Inequality: Causes and Consequences (Research in Labor Economics, Volume 43) Emerald Group Publishing Limited, 43, Møen, J., K. G. Salvanes, and E. Ø. Sørensen (2003): Documentation of the Linked Empoyer Employee Data Base at the Norwegian School of Economics, Working paper, The Norwegian School of Economics. Morissette, R., P. C. W. Chan, and Y. Lu (2015): Wages, Youth Employment, and School Enrollment: Recent Evidence from Increases in World Oil Prices, Journal of Human Resources, 50(1), Noreng, O. (1980): The Economics and Politics of Oil and Gas. Routledge Library Editions, London. NOU (1972): Tilleggsberegninger Ajourførte pris-og inntektsprognoser for 1972 og 1973 fra Det tekniske beregningsutvalg, Rapport 1972/2, Regjeringen Korvald. 29

31 Nybom, M., and J. Stuhler (2013): Interpreting Trends in Intergenerational Income Mobility, IZA Discussion Papers 7514, Institute for the Study of Labor (IZA). (2016): Biases in Standard Measures of Intergenerational Income Dependence, Journal of Human Resources. Pekkarinen, T., K. Salvanes, and M. Sarvimäki (2017): The Evolution of Social Mobility: Norway over the 20th Century, Scandinavian Journal of Economics, 119(1), Solon, G. (1999): Intergenerational Mobility in the Labor Market, in Handbook of Labor Economics, ed. by O. Ashenfelter, and D. Card, vol. 3 of Handbook of Labor Economics, chap. 29, pp Elsevier. 30

32 8 Tables and Figures Table 1: Descriptive Statistics Cohorts Cohorts Low Oil High Oil Low Oil High Oil Regions Regions Regions Regions 1st generation Age at son s birth (4.6) (4.6) Earnings at ages , ,799 63,821 59,092 (114,985) (109,003) (34,908) (28,456) Years of education (2.97) (2.75) % academic high school nd generation Earnings at age , , , ,469 (173,421) (153,302) (83,349) (82,931) Years of education (2.85) (2.72) (3.13) (3.07) % academic high school % employed in oil sector % received disability insurance rd generation Earnings at age , ,749 (168,680) (175,810) % academic high school IQ measure Number of father-son couples 1st and 2nd generations 70,239 15,688 13,176 3,054 2nd and 3rd generations 92,784 24,210 Note: Means and standard deviations in parentheses. Individuals are classified as employed in the oil sector if they work in crude petroleum and natural gas production, petroleum refining, manufacturing of products of petroleum and coal, manufacturing of machinery, manufacturing of transport equipment, and construction other than building construction, which includes oil well drilling, in the definition of oil sector jobs. For birth cohorts , we use industry codes from 1986; for birth cohorts , we define a worker as employed in the oil sector if his main occupation between ages 36 and 41 is in an oil-related industry. The earnings of fathers of birth cohorts are imputed following Pekkarinen, Salvanes, and Sarvimäki (2016) (see Section 4 for a detailed discussion). For the third generation, we observe earnings at age 30 only for individuals born until 1985 (46% of all sons born to birth cohorts ) and we observe whether individuals completed academic education for those born until 1991 (78% of all sons born to birth cohorts ). 31

33 Table 2: Rank-Rank Regressions Cohorts Cohorts (i) (ii) Fathers rank 0.228*** 0.235*** (0.013) (0.004) High oil* fathers rank *** (0.023) (0.011) Number of observations 16,230 85,927 R-squared Note: Each column is from a separate regression of the rank of the son in the son s birth cohort s earnings distribution on the father s rank in the earnings distribution of fathers with sons born in the same year. Robust standard errors adjusted for clustering at the municipality level are shown in parentheses. We include birth cohorts from 1932 to 1933 in Column (i) and birth cohorts from 1952 to 1957 in Column (ii). All specifications include a full set of cohort and local labor market fixed effects. In Column (ii), we control in addition for father s age at birth. Table 3: Probability of Sons Being in Earnings Quintile Given the Earnings Quintile of the Father, Cohorts Panel A: Low-Oil Local Labor Market Regions Sons in Sons in Sons in Sons in Sons in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Fathers in 1st quintile Fathers in 2nd quintile Fathers in 3rd quintile Fathers in 4th quintile Fathers in 5th quintile Panel B: High-Oil Local Labor Market Regions Sons in Sons in Sons in Sons in Sons in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Fathers in 1st quintile Fathers in 2nd quintile Fathers in 3rd quintile Fathers in 4th quintile Fathers in 5th quintile Note: Each cell indicates the percentage of sons growing up in a given earnings quintile in their fathers earnings distribution who end up in a specific earnings quintile in their cohort s earnings distribution. Son s earnings are ranked in the son s birth cohort s earnings distribution. Father s earnings are ranked in the earnings distribution of fathers with sons born in the same year. We include birth cohorts from 1952 to The figures in italics indicate that the difference between the low- and high-oil regions is significantly different from zero at the 5% level. 32

34 Table 4: Probability of Sons Being in Earnings Quintile Given the Earnings Quintile of their Fathers, Cohorts Panel A: Low-Oil Local Labor Market Regions Sons in Sons in Sons in Sons in Sons in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Fathers in 1st quintile Fathers in 2nd quintile Fathers in 3rd quintile Fathers in 4th quintile Fathers in 5th quintile Panel B: High-Oil Local Labor Market Regions Sons in Sons in Sons in Sons in Sons in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Fathers in 1st quintile Fathers in 2nd quintile Fathers in 3rd quintile Fathers in 4th quintile Fathers in 5th quintile Note: Each cell indicates the percentage of sons growing up in a given earnings quintile in their fathers earnings distribution who end up in a specific earnings quintile in their cohort s earnings distribution. Son s earnings are ranked in the son s birth cohort s earnings distribution. Father s earnings are ranked in the earnings distribution of fathers with sons born in the same year. We include birth cohorts from 1932 to The figures in italics indicate that the difference between the low- and high-oil regions is significantly different from zero at the 5% level. 33

35 Table 5: Rank-Education Regressions Cohorts Cohorts (i) (ii) Fathers rank *** ( ) ( ) Fathers rank *** *** ( ) ( ) High oil* fathers rank ( ) ( ) High oil* fathers rank ( ) ( ) Number of observations 16,226 85,997 R-squared Note: Each column is from a separate regression of an indicator variable whether the son completed academic high school before turning 30, on the father s rank in the earnings distribution of fathers and a quadratic function of the father s rank. Father s earnings are ranked in the earnings distribution of fathers with sons born in the same year. Robust standard errors adjusted for clustering at the municipality level are shown in parentheses. We include birth cohorts from 1932 to 1933 in Column (i) and birth cohorts from 1952 to 1957 in Column (ii). All specifications include a full set of cohort and local labor market fixed effects. In Column (ii), we control for father s age at birth. Table 6: Returns to Academic Education Cohorts Cohorts (i) (ii) Academic high school 0.349*** 0.447*** (0.011) (0.010) Academic high school*high oil *** (0.032) (0.018) Number of observations 16,092 82,986 R-squared Number of observations 16,041 82,986 R-squared Note: Each column is from a separate regression of the log average earnings at ages years for dummies of educational attainment at age 36. The regression equation includes an indicator variable for having completed academic education (either an academic high school and/or college degree) and an interaction term of the academic education indicator and an indicator for being born in high-oil region. The excluded categories are workers who completed vocational high school or did not complete academic high school education. Robust standard errors clustered at the municipality level are shown in parentheses. We include birth cohorts from 1932 to 1933 in Column (i) and birth cohorts from 1952 to 1957 in Column (ii). All specifications include a full set of cohort and local labor market fixed effects. 34

36 Table 7: Inequality Measures for the Earnings Distribution for Different Male Cohorts in Different Oil Regions Cohort Region p90/p10 p90/p50 p10/p50 p75/p25 Gini (i) (ii) (iii) (iv) (v) Low Oil High Oil Low Oil High Oil Note: Each row displays different indexes of earnings inequality for different cohorts in different oil regions. We include earnings of all the male workers who are in our main analysis. Columns (i) to (iv) present the ratio between different percentiles of the earnings distribution. Columns (i) and (iv) are measures of the overall dispersion (ratio of the 90th and the 10th percentile and ratio of the 75th and the 25th percentile, respectively). Column (ii) is a measure of dispersion in the top part of the earnings distribution (ratio of the 90th and the 50th percentile). Column (iii) is a measure of dispersion in the bottom part (ratio of the 50th and the 10th percentile). Column (v) reports the Gini index of the earnings distributions. Earnings are defined as the average annual earnings a person earned between ages 36 and 41. Table 8: Rank-Rank Regressions for the Relationship Sons Grandfathers and Sons Fathers, Sons of Cohorts Sons Grandfathers (i) Sons Fathers (ii) Grandfathers rank 0.053*** (0.006) Fathers rank 0.166*** (0.007) High oil* grandfathers rank *** (0.014) High oil* fathers rank *** (0.014) Number of observations 39,684 39,684 R-squared Note: Each column is from a separate regression of the percentile rank of the son on the grandfathers percentile rank (Column i) and on the father s percentile rank (Column ii). Sons earnings are ranked in the earnings distribution of other children with fathers born in the same year, grandfathers earnings are ranked in the earnings distribution of grandfathers with sons born in the same year, and fathers earnings are ranked in the earnings distribution of other men born in the same year. Robust standard errors adjusted for clustering at the municipality level are shown in parentheses. All specifications include a full set of fathers cohort and local labor market fixed effects and father s age at son s birth. 35

37 Table 9: Number and Characteristics of Movers Across Regions Low-to-High Low High-to-Low High Movers Stayers Movers Stayers Fraction Avg father 256, , , ,251 earnings (50 55) Cohort Avg father education Avg earnings 232, , , ,827 age Proportion with academic degree Avg age at movement Fraction Avg father 224, , , ,449 earnings (50 55) Cohort Avg father education Avg earnings 293, , ,276 age Proportion with academic degree Avg age at movement Notes: A person is considered to move across regions if he is registered as living in another region at least once between 18 and 41 years of age than the region of birth. Proportions are to be interpreted with respect to the total population born in a specific area. That is, the proportion of movers migrating from the low-oil region to the high-oil region is the share of those who migrate to the high-oil region among those born in the low-oil region. 36

38 Table 10: Robustness Analysis Place of Excluding Place Excluding birth migrants of residence Oslo (i) (ii) (iii) (iv) Fathers rank 0.235*** 0.230*** 0.222*** 0.229*** (0.004) (0.005) (0.005) (0.000) High oil* fathers rank *** *** ** ** (0.011) (0.013) (0.010) (0.010) Number of observations 85,927 67,426 85,819 69,812 R-squared Notes: Each column is from a separate regression of the rank of the son in the son s birth cohort s earnings distribution on the father s rank in the earnings distribution of fathers with sons born in the same year. Robust standard errors adjusted for clustering at the municipality level are shown in parentheses. All specifications include a full set of cohort and local labor market fixed effects and father s age at birth. We include birth cohorts In Column (i), we assign individuals to an oil region based on their place of birth. In Column (ii), we exclude individuals migrating across oil regions. In Column (iii), we assign individuals to an oil region based on their place of residence at age 36. Note that the number of observations is not the same in Column (i) and Column (iii) because of the number of individuals born and living in regions where oil employment is between 7.5% and 10%, who are excluded from the regression, is different. Column (iv) excludes all individuals born in Oslo. 37

39 Table 11: Son s Probability of Disability Insurance Receipt Conditional on Son s and Father s Earnings Quintile Panel A: Low-Oil Local Labor Market Regions Sons in Sons in Sons in Sons in Sons in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Fathers in 1st quintile Fathers in 2nd quintile Fathers in 3rd quintile Fathers in 4th quintile Fathers in 5th quintile Panel B: High-Oil Local Labor Market Regions Sons in Sons in Sons in Sons in Sons in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Fathers in 1st quintile Fathers in 2nd quintile Fathers in 3rd quintile Fathers in 4th quintile Fathers in 5th quintile Note: Each cell indicates the percentage of sons enrolling in disability insurance given their earnings quintile and the earnings quintile of their father. Son s earnings are ranked in the son s birth cohort s earnings distribution. Father s earnings are ranked in the earnings distribution of fathers with sons born in the same year. We include birth cohorts from 1952 to The figures in italics indicate that the difference between the low- and high-oil regions is significantly different from zero at the 5% level. 38

40 Figure 1: Trends in Earnings per Capita, by Local Labor Market, Income per capita in NOK (deviation from the yearly average) Year Low Oil High Oil Notes: The figure plots the yearly regional deviation from the average national earnings per capita in the high- and low-oil region. Data source: own calculations based on tax data available for the years: 1951, 1954, 1957, 1961, 1965, 1968, 1971, 1972, 1973, 1974, 1975, 1976, 1977, 1978, 1979, 1980, and

41 Figure 2: Proportion of Individuals Finishing Academic High School by Birth Cohort Proportion with academic high school degree or college degree, males Year of birth Low oil High oil Notes: The figure plots the proportion of men in each birth cohort who finish academic high school. An academic high school degree is required to enroll in university or college. Figure 3: Proportion of Individuals Finishing Vocational High School by Birth Cohort Proportion with vocational high school degree, males Year of birth Low oil High oil Notes: The figure plots the proportion of men in each birth cohort who finish vocational high school. 40

42 Figure 4: Employment in the Oil Industry in 1980 by Local Labor Market (.1,.3] (.075,.1] [0,.075] Notes: The figure plots the proportion of workers employed in the oil industry in 1980 in each of the 46 local labor market. Local labor markets are aggregations of municipalities, based on commuting patterns (Bhuller, 2009). 41

43 Figure 5: Association between Sons and Fathers Percentile Ranks by Local Labor Markets, Cohorts Average income rank, sons Fathers rank (among other fathers with sons in the same cohort) High Oil High Oil 95% CI Low Oil Low Oil Notes: These figures present nonparametric binned scatterplots of the relationship between sons percentile rank in the earnings distribution versus fathers percentile rank in the fathers earnings distribution in the high- and low-oil regions. We include birth cohorts Son s earnings are ranked in the son s birth cohort s earnings distribution. Father s earnings are ranked in the earnings distribution of fathers with sons born in the same year. Figure 6: Association between Sons and Fathers Percentile Ranks by Local Labor Markets, Cohorts Average income rank, sons Fathers rank (among other fathers with sons in the same cohort) High Oil High Oil 95% CI Low Oil Low Oil Notes: These figures present nonparametric binned scatterplots of the relationship between sons percentile rank in the earnings distribution versus the fathers percentile rank in the fathers earnings distribution in the high- and low-oil regions. We include birth cohorts Son s earnings are ranked in the son s birth cohort s earnings distribution. Father s earnings are ranked in the earnings distribution of fathers with sons born in the same year. 42

44 Figure 7: Association between Fathers Percentile Ranks and Son s Education, Cohorts Proportion of sons with academic high school degree or college degree Fathers rank (among other fathers with sons in the same cohort) High Oil High Oil 95% CI Low Oil Low Oil Notes: These figures present nonparametric binned scatterplots of the relationship between sons likelihood of finishing an academic high school degree versus fathers percentile rank in the fathers earnings distribution in the high- and low-oil regions. We include birth cohorts Father s earnings are ranked in the earnings distribution of fathers with sons born in the same year. Figure 8: Association between Fathers Percentile Ranks and Son s Education, Cohorts Proportion of sons with academic high school degree or college degree Fathers rank (among other fathers with sons in the same cohort) High Oil High Oil 95% CI Low Oil Low Oil Notes: These figures present nonparametric binned scatterplots of the relationship between sons likelihood of finishing an academic high school degree versus fathers percentile rank in the fathers earnings distribution in the high- and low-oil regions. We include birth cohorts Father s earnings are ranked in the earnings distribution of fathers with sons born in the same year. 43

45 Figure 9: Earnings Distribution by Cohorts and Regions Density e e e Average earnings between 36 and 41, sons Cohorts 32 33, Low oil Cohorts 52 57, Low oil Cohorts 32 33, High oil Cohorts 52 57, High oil Notes: These figures present kernel density plots of son s average earnings ages years. The figure is based on the earnings of the and birth cohorts. 44

46 Figure 10: Earnings Distribution by Type of Education and Regions Density e e Average earnings between 36 and 41 Vocational, Low oil Academic, Low oil Vocational, High oil Academic, High oil Notes: These figures present kernel density plots of son s average earnings ages years. The figure is based on the earnings of the birth cohorts. Figure 11: Earnings Distribution by Industry and Regions for Vocationally Trained Individuals Density e e e Average Earnings between 36 and 41 No Oil Sector, Low Oil Area No Oil Sector, High Oil Area Oil Sector, Low Oil Area Oil Sector, High Oil Area Notes: These figures present kernel density plots of vocationally educated sons average earnings ages years. The figure is based on the earnings of the birth cohorts. 45

47 Figure 12: Earnings Distribution by Industry and Regions for Academically Trained Individuals Density e e Average Earnings between 36 and 41 No Oil Sector, Low Oil Area No Oil Sector, High Oil Area Oil Sector, Low Oil Area Oil Sector, High Oil Area Notes: These figures present kernel density plots of academically educated sons average earnings ages years. The figure is based on the earnings of the birth cohorts. Figure 13: Association between Sons and Grandfathers Percentile Ranks by Local Labor Markets, Sons of Cohorts Average income rank, grandsons Grandfathers rank (among other grandfathers with sons in the same cohort) High Oil High Oil 95% CI Low Oil Low Oil Notes: These figures present nonparametric binned scatterplots of the relationship between sons percentile rank in the earnings distribution versus grandfathers percentile rank in the grandfathers earnings distribution in the highand low-oil regions. The figure is based on the sons of the birth cohorts. The grandfathers are the fathers of the birth cohorts. Son s earnings are ranked in the earnings distribution of other sons with fathers born in the same year. Grandfather s earnings are ranked in the earnings distribution of grandfathers with their sons born in the same year. Sons earnings are observed at age

48 Figure 14: Association between Sons and Fathers Percentile Ranks by Local Labor Markets, Sons of Cohorts Average income rank, grandsons Fathers rank High Oil High Oil 95% CI Low Oil Low Oil Notes: These figures present nonparametric binned scatterplots of the relationship between sons percentile rank in the earnings distribution versus fathers percentile rank in the fathers earnings distribution in the high- and low-oil regions. The figure is based on the sons of the birth cohorts. Fathers belong to the birth cohorts. Son s earnings are ranked in the earnings distribution of other sons with fathers born in the same year. Father s earnings are ranked in the father s birth cohort s earnings distribution. Son s earnings are observed at age

49 Figure 15: Average Earnings Rank at Age 30, Sons of Cohorts , Low-Oil Region Notes: This figure plots the average earnings percentile rank (expressed in 20 percentiles) for the sons of the cohorts in the low-oil region as a function of the father s and grandfather s earnings rank. The sample includes sons born before

50 Figure 16: Average Earnings Rank at Age 30, Sons of Cohorts , High-Oil Region Notes: This figure plots the average earnings percentile rank (expressed in 20 percentiles) for the sons of the cohorts in the high-oil region as a function of the earnings ranks of fathers and grandfathers The sample includes sons born before

51 Figure 17: Probability of Attending Academic High School, Sons of Cohorts , Low-Oil Region Notes: This figure plots the probability of completing academic high school or college for the sons of the cohorts in the low-oil region as a function of the earnings ranks of fathers and grandfathers The sample includes sons born before

52 Figure 18: Probability of Attending Academic High School, Sons of Cohorts , High-Oil Region Notes: This figure plots the probability of completing academic high school or college for the sons of the cohorts in the high-oil region as a function of the earnings ranks of fathers and grandfathers The sample includes sons born before

53 A Appendix Table A1: Rank-Rank Regressions for Different Cohorts Cohorts Cohorts (i) (ii) (iii) (iv) (v) (vi) (vii) Father s rank 0.235*** 0.233*** 0.234*** 0.234*** 0.230*** 0.223*** 0.221*** (0.004) (0.005) (0.004) (0.004) (0.003) (0.003) (0.003) Father s rank * *** *** *** *** *** *** *** High oil (0.011) (0.011) (0.011) (0.011) (0.010) (0.008) (0.006) Number of observations 85,922 96, , , , , ,277 R-squared Note: Each column is from a separate regression of the rank of the son in the son s birth cohort s earnings distribution on the father s rank in the earnings distribution of fathers with sons born in the same year. Robust standard errors adjusted for clustering at the municipality level are shown in parentheses. We include birth cohorts Column (i) includes birth cohorts , Column (ii) includes birth cohorts 1954=-1959, Column (iii) includes birth cohorts , Column (iv) includes birth cohorts , Column (v) includes birth cohorts , Column (vi) includes birth cohorts , and Column (vii) includes birth cohorts All specifications include a full set of cohort and local labor market fixed effects and father s age at birth. 52

54 Table A2: Probability of Sons Being in Earnings Quintile Given the Earnings Quintile of the Father, Cohorts , Excluding Movers Panel A: Low-Oil Local Labor Market Region Sons in Sons in Sons in Sons in Sons in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Fathers in 1st quintile Fathers in 2nd quintile Fathers in 3rd quintile Fathers in 4th quintile Fathers in 5th quintile Panel B: High-Oil Local Labor Market Region Sons in Sons in Sons in Sons in Sons in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Fathers in 1st quintile Fathers in 2nd quintile Fathers in 3rd quintile Fathers in 4th quintile Fathers in 5th quintile Note: Each cell indicates the percentage of sons growing up in a given earnings quintile in their fathers earnings distribution who end up in a specific earnings quintile in their cohort s earnings distribution. Son s earnings are ranked in the son s birth cohort s earnings distribution. Father s earnings are ranked in the earnings distribution of fathers with sons born in the same year. We include birth cohorts Individuals who move across oil regions are excluded. The figures in italics indicate that the difference between the low- and high-oil regions is significantly different from zero at the 5% level. 53

55 Table A3: Probability of Sons Being in Earnings Quintile Given the Earnings Quintile of the Father, Cohorts , Assigning Workers to Their Place of Residence at Age 36 Panel A: Low-Oil Local Labor Market Region Sons in Sons in Sons in Sons in Sons in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Fathers in 1st quintile Fathers in 2nd quintile Fathers in 3rd quintile Fathers in 4th quintile Fathers in 5th quintile Panel B: High-Oil Local Labor Market Region Sons in Sons in Sons in Sons in Sons in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Fathers in 1st quintile Fathers in 2nd quintile Fathers in 3rd quintile Fathers in 4th quintile Fathers in 5th quintile Note: Each cell indicates the percentage of sons growing up in a given earnings quintile in their fathers earnings distribution who end up in a specific earnings quintile in their cohort s earnings distribution. Son s earnings are ranked in the son s birth cohort s earnings distribution. Father s earnings are ranked in the earnings distribution of fathers with sons born in the same year. We include birth cohorts We assign individuals to an oil region based on their place of residency at age 36. The figures in italics indicate that the difference between the low- and high-oil regions is significantly different from zero at the 5% level. 54

56 Table A4: Average Earnings Quintile of Sons Given the Earnings Quintiles of their Father and Grandfather, Sons of Cohorts Panel A: Low-Oil Local Labor Market Region Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Panel B: High-Oil Local Labor Market Region Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Note: Each cell indicates the average earnings quintile of the sons of birth cohorts given the earnings quintiles of their father and grandfather. Father s earnings are ranked in the father s birth cohort s earnings distribution. Grandfather s earnings are ranked in the earnings distribution of grandfathers with sons born in the same year. Sons earnings (measured at age 30) are ranked in the earnings distribution of sons with fathers in the same birth cohort. We include sons born before The figures in italics indicate that the difference between the low- and high-oil regions is significantly different from zero at the 5% level. 55

57 Table A5: Proportion of Sons with an Academic Education Given the Earnings Quintiles of their Father and Grandfather, Sons of Cohorts Panel A: Low-Oil Local Labor Market Region Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Panel B: High-Oil Local Labor Market Region Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Note: Each cell indicates the percentage of sons completing academic high school or obtaining a university degree given the earnings quintiles of their father and grandfather. Father s earnings are ranked in the father s birth cohort s earnings distribution. Grandfather s earnings are ranked in the earnings distribution of grandfathers with sons born in the same year. We include the sons of birth cohorts born before The figures in italics indicate that the difference between the low- and high-oil regions is significantly different from zero at the 5% level. 56

58 Table A6: Proportion of Fathers with an Academic Education Given their Earnings Quintile and the Earnings Quintile of their Grandfather, Cohorts Panel A: Low-Oil Local Labor Market Region Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Panel B: High-Oil Local Labor Market Region Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Note: Each cell indicates the percentage of fathers completing academic high school or obtaining a university degree given their earnings quintile and the earnings quintile of their grandfather. Father s earnings are ranked in the father s birth cohort s earnings distribution. Grandfather s earnings are ranked in the earnings distribution of grandfathers with sons born in the same year. We include birth cohorts The figures in italics indicate that the difference between the low- and high-oil regions is significantly different from zero at the 5% level. 57

59 Table A7: Proportion of Fathers Spouses with an Academic Education Given the Earnings Quintiles of the Father and Grandfathers, Cohorts Panel A: Low-Oil Local Labor Market Region Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Panel B: High-Oil Local Labor Market Region Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Note: Each cell indicates the percentage of wives completing academic high school or obtaining a university degree given the earnings quintiles of their husbands (fathers) and their husbands fathers (grandfathers). Father s earnings are ranked in the father s birth cohort s earnings distribution. Grandfathers earnings are ranked in the earnings distribution of grandfathers with sons born in the same year. We include wives of birth cohorts The figures in italics indicate that the difference between the low- and high-oil regions is significantly different from zero at the 5% level. 58

60 Table A8: Rank-Rank Regressions for the Relationships Sons Grandfathers and Sons Fathers, Sons of Cohorts , Experience-adjusted Earnings Sons Grandfathers (i) Sons Fathers (ii) Grandfathers rank 0.048*** (0.006) Fathers rank 0.151*** (0.006) High oil* grandfathers rank *** (0.015) High oil* fathers rank *** (0.014) Number of observations 39,683 39,683 R-squared Note: Each column is from a separate regression of the percentile rank of the sons on the grandfathers percentile rank (Column i) and on the father s percentile rank (Column ii). Sons earnings are adjusted for their experience on the labor market and are ranked in the (adjusted) earnings distribution of other sons with fathers born in the same year, grandfathers earnings are ranked in the earnings distribution of grandfathers with sons born in the same year, and fathers earnings are ranked in the earnings distribution of other men born in the same year. Robust standard errors adjusted for clustering at the municipality level are shown in parentheses. All specifications include a full set of fathers cohort and local labor market fixed effects and father s age at son s birth. Son s earnings are observed at age 30. We predict experience-adjusted earnings at age 30 based on a second-order polynomial function of experience. 59

61 Table A9: Average Earnings Quintile of Sons Given the Earnings Quintiles of their Father and Grandfather, Sons of Cohorts , Experience-adjusted Earnings Panel A: Low-Oil Local Labor Market Regions Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Panel B: High-Oil Local Labor Market Regions Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Note: Each cell indicates the average earnings quintile of the sons of birth cohort given the earnings quintiles of their father and grandfather. Father s earnings are ranked in the father s birth cohort s earnings distribution. Grandfather s earnings are ranked in the earnings distribution of grandfathers with sons born in the same year. Sons earnings are adjusted for experience on the labor market and are ranked in the (adjusted) earnings distribution of sons with fathers in the same birth cohort. We include sons born before Son s earnings are observed at age 30. We predict experience-adjusted earnings at age 30 based on a second-order polynomial function of experience. The figures in italics indicate that the difference between the low- and high-oil regions is significantly different from zero at the 5% level. 60

62 Table A10: Average IQ of Sons Given the Earnings Quintiles of their Father and Grandfather, Sons of Cohorts Panel A: Low-Oil Local Labor Market Regions Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Panel B: High-Oil Local Labor Market Regions Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Note: Each cell indicates the IQ scores of sons given the earnings quintiles of their father and grandfather. Father s earnings are ranked in the father s birth cohort s earnings distribution. Grandfather s earnings are ranked in the earnings distribution of grandfathers with sons born in the same year. We include sons of birth cohorts born before IQ scores are taken at age 19 by all males (military conscription) and are measures on a stanine scale (between 1 and 9). The figures in italics indicate that the difference between the low- and high-oil regions is significantly different from zero at the 5% level. 61

63 Table A11: Rank-Rank Regressions for the Relationships Daughters Grandfathers and Daughters Fathers, Daughters of Cohorts Daughters Grandfathers (i) Daughters Fathers (ii) Grandfathers rank 0.083*** (0.007) Fathers rank 0.168*** (0.007) High oil* grandfathers rank (0.012) High oil* fathers rank *** (0.014) Number of observations 37,752 37,752 R-squared Note: Each column is from a separate regression of the percentile rank of the daughter on the grandfathers percentile rank (Column i) and on the father s percentile rank (Column ii). Daughters earnings are ranked in the earnings distribution of other daughters with fathers born in the same year, grandfathers earnings are ranked in the earnings distribution of grandfathers with sons born in the same year, and fathers earnings are ranked in the earnings distribution of other men born in the same year. Robust standard errors adjusted for clustering at the municipality level are shown in parentheses. All specifications include a full set of fathers cohort and local labor market fixed effects and father s age at daughter s birth. 62

64 Table A12: Average Earnings Quintile of Daughters Given the Earnings Quintiles of their Father and Grandfather, Daughters of Cohorts Panel A: Low-Oil Local Labor Market Region Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Panel B: High-Oil Local Labor Market Region Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Note: Each cell indicates the average earnings quintile of the daughters of birth cohort given the earnings quintiles of their father and grandfather. Father s earnings are ranked in the father s birth cohort s earnings distribution. Grandfather s earnings are ranked in the earnings distribution of grandfathers with sons born in the same year; daughters earnings (measured at age 30) are ranked in the earnings distribution of daughters with fathers in the same birth cohort. We include the daughters born before The figures in italics indicate that the difference between the low- and high-oil regions is significantly different from zero at the 5% level. 63

65 Table A13: Proportion of Daughters with an Academic Education Given the Earnings Quintiles of their Father and Grandfather, Daughters of Cohorts Panel A: Low-Oil Local Labor Market Region Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Panel B: High-Oil Local Labor Market Region Fathers in Fathers in Fathers in Fathers in Fathers in 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile Grandfathers in 1st quintile Grandfathers in 2nd quintile Grandfathers in 3rd quintile Grandfathers in 4th quintile Grandfathers in 5th quintile Note: Each cell indicates the percentage of daughters completing academic high school or obtaining a university degree given the earnings quintiles of their father and grandfather. Father s earnings are ranked in the father s birth cohort s earnings distribution. Grandfather s earnings are ranked in the earnings distribution of grandfathers with sons born in the same year. We include the daughters of birth cohorts from 1952 to 1957 born before The figures in italics indicate that the difference between the low- and high-oil regions is significantly different from zero at the 5% level. 64

66 Figure A1: Licensed Blocks on the Norwegian Continental Shelf Notes: The map shows the licensing position from the first round of licenses issued by the Norwegian state in Source: Norwegian Petroleum Directorate. 65

67 Figure A2: Oil and Gas Discoveries until 2015 Notes: The map shows all offshore oil and gas discoveries until 2015 (in blue). The three largest Norwegian cities Oslo, Bergen, and Stavanger are marked in red. The map also displays the location of the Ekofisk oil field, which was the first oil discovery on the Norwegian continental shelf. Source: Norwegian Petroleum Directorate. 66

68 Figure A3: Main Oil and Gas Industry Supply Bases in 2016 Notes: The map shows all main oil and gas industry supply bases operating in These supply bases include Tananger and Dusavik (close to Stavanger), Sotra and Mongstad (close to Bergen), Florø, Kristiansund, Sandnessjøen, and Hammerfest. Source: Norwegian Petroleum Directorate. 67

69 Figure A4: Distribution of Age at Birth of the First Child, Cohorts Percent Age at first child birth Low Oil High Oil Notes: This figure plots the distributions of age at first child for workers born in high- and low-oil regions in the years

70 Figure A5: Association between Sons and Grandfathers Percentile Ranks by Local Labor Markets, Sons of Cohorts , Experience-adjusted Earnings Average income rank, grandsons Grandfathers rank (among other grandfathers with sons in the same cohort) High Oil High Oil 95% CI Low Oil Low Oil Notes: This figure presents nonparametric binned scatterplots of the relationship between sons percentile rank in the earnings distribution versus grandfathers percentile rank in the grandfathers earnings distribution in the highand low-oil regions. The figure is based on the sons of the birth cohorts. The grandfathers are the fathers of the birth cohorts. Son s earnings are adjusted for experience on the labor market and are ranked in the (adjusted) earnings distribution of other sons with fathers born in the same year. Grandfather s earnings are ranked in the earnings distribution of grandfathers with their sons born in the same year. Son s earnings are observed at age 30. We predict experience-adjusted earnings at age 30 based on a second-order polynomial function of experience. 69

71 Figure A6: Association between Sons and Fathers Percentile Ranks by Local Labor Markets, Sons of Cohorts , Experience-adjusted Earnings Average income rank, grandsons Fathers rank High Oil High Oil 95% CI Low Oil Low Oil Notes: This figure presents nonparametric binned scatterplots of the relationship between sons percentile rank in the earnings distribution versus fathers percentile rank in the fathers earnings distribution in the high- and low-oil regions. The figure is based on the sons of the birth cohorts. Fathers belong to the birth cohorts. Sons earnings are adjusted for experience on the labor market and are ranked in the (adjusted) earnings distribution of other sons with fathers born in the same year. Father s earnings are ranked in the father s birth cohort s earnings distribution. Son s earnings are observed at age 30. We predict experience-adjusted earnings at age 30 based on a second-order polynomial function of experience. 70

72 Figure A7: Average Earnings Rank at Age 30, Sons of Cohorts , Low-Oil Region, Experience-adjusted Earnings Notes: This figure plots the average experience-adjusted earnings percentile rank (expressed in 20 percentiles) for the sons of the cohorts in the low-oil region as a function of the earnings ranks of fathers and grandfathers The sample includes sons born before Son s earnings are observed at age 30. We predict experience-adjusted earnings at age 30 based on a second-order polynomial function of experience. 71

73 Figure A8: Average Earnings Rank at Age 30, Sons of Cohorts , High-Oil Region, Experience-adjusted Earnings Notes: This figure plots the average experience-adjusted earnings percentile rank (expressed in 20 percentiles) for the sons of the cohorts in the high-oil region as a function of the earnings ranks of fathers and grandfathers The sample includes sons born before Son s earnings are observed at age 30. We predict experience-adjusted earnings at age 30 based on a second-order polynomial function of experience. 72

74 Figure A9: Average IQ, Sons of Cohorts , Low-Oil Region Notes: This figure plots the average IQ score for the sons as a function of the earnings rank of their fathers and grandfathers in the low-oil region. We include sons of birth cohorts born before IQ scores are taken at age 19 by all males (military conscription) and are measures on a stanine scale (between 1 and 9). 73

75 Figure A10: Average IQ, Sons of Cohorts , High-Oil Region Notes: This figure plots the average IQ score for the sons as a function of the earnings rank of their fathers and grandfathers in the high-oil region. We include sons of birth cohorts born before IQ scores are taken at age 19 by all males (military conscription) and are measured on a stanine scale (between 1 and 9). 74

76 Figure A11: Association between Daughters and Grandfathers Percentile Ranks by Local Labor Markets, Daughters of Cohorts Average income rank, grandsons Grandfathers rank (among other grandfathers with sons in the same cohort) High Oil High Oil 95% CI Low Oil Low Oil Notes: These figures present nonparametric binned scatterplots of the relationship between daughters percentile rank in the earnings distribution versus grandfathers percentile rank in the grandfathers earnings distribution in the high- and low-oil regions. The figure is based on the daughters of the birth cohorts. The grandfathers are the fathers of the birth cohorts. Daughter s earnings are ranked in the earnings distribution of other daughters with fathers born in the same year. Grandfather s earnings are ranked in the earnings distribution of grandfathers with their sons born in the same year. Daughter s earnings are observed at age

77 Figure A12: Association between Daughters and Fathers Percentile Ranks by Local Labor Markets, Daughters of Cohorts Average income rank, grandsons Fathers rank High Oil High Oil 95% CI Low Oil Low Oil Notes: These figures present nonparametric binned scatterplots of the relationship between daughters percentile rank in the earnings distribution versus fathers percentile rank in the fathers earnings distribution in the high- and low-oil regions. The figure is based on the daughters of the birth cohorts. Fathers belong to the birth cohorts. Daughter s earnings are ranked in the earnings distribution of other daughters with fathers born in the same year. Father s earnings are ranked in the father s birth cohort s earnings distribution. Daughter s earnings are observed at age

78 Figure A13: Average Earnings Rank at Age 30, Daughters of Cohorts , Low-Oil Region Notes: This figure plots the average earnings percentile rank (expressed in 20 percentiles) for the daughters of the cohorts in the low-oil region as a function of the earnings ranks of fathers and grandfathers The sample includes daughters born before

79 Figure A14: Average Earnings Rank at Age 30, Daughters of Cohorts , High-Oil Region Notes: This figure plots the average earnings percentile rank (expressed in 20 percentiles) for the daughters of the cohorts in the high-oil region as a function of the earnings ranks of fathers and grandfathers The sample includes daughters born before