Employment trends in Norway

Ellingsen, Nicolai; Fosso, Luca; Galaasen, Sigurd Mølster Research Report Employment trends in Norway Staff Memo, No. 1/2024 Provided in Cooperation with: Norges Bank, Oslo Suggested Citation: Ellingsen, Nicolai; Fosso, Luca; Galaasen, Sigurd Mølster (2024) : Employment trends in Norway, Staff Memo, No. 1/2024, ISBN 978-82-8379-307-9, Norges Bank, Oslo, https://hdl.handle.net/11250/3170067 This Version is available at: https://hdl.handle.net/10419/310421 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/ Staff Memo Employment trends in Norway Authors: Nicolai Ellingsen Luca Fosso Sigurd Mølster Galaasen 1 | 2024 Norges Bank Staff Memo 1 Staff Memos present reports and documentation written by staff members and affiliates of Norges Bank, the central bank of Norway. Views and conclusions expressed in Staff Memos should not be taken to represent the views of Norges Bank. © 2024 Norges Bank The text may be quoted or referred to, provided that due acknowledgement is given to source. Staff Memo inneholder utredninger og dokumentasjon skrevet av Norges Banks ansatte og andre forfattere tilknyttet Norges Bank. Synspunkter og konklusjoner i arbeidene er ikke nødvendigvis representative for Norges Banks. © 2024 Norges Bank Det kan siteres fra eller henvises til dette arbeid, gitt at forfatter og Norges Bank oppgis som kilde. ISSN 1504-2596 (online) ISBN 978-82-8379-307-9 (online) Employment trends in Norway* Nicolai Ellingsen † Norges Bank Luca Fosso ‡ European Central Bank Sigurd Mølster Galaasen § Norges Bank January 17, 2024 Abstract This paper outlines the recently developed method for assessing the trend level in employment rates adopted by Norges Bank. The approach employs a Bayesian VAR to decompose disaggregated employment data into trend and cyclical components, using quarterly labor market data on 30 demographic sub-groups. Applied to the time period 1984-2019, we show that the estimated trend picks up known historical factors contributing to slow-moving employment dynamics. Additionally, the cyclical employment component shows strong correlation with the output gap estimate for Norway. 1 Introduction Ensuring that a large part of the population is employed is a fundamental objective of economic policy. To translate this goal into policy actions, it is necessary to both define what it means to have a high rate of employment and to measure how this objective changes over time. This is a challenging task, as the observed level of employment is the outcome of a wide range of factors that vary in terms of their observability, predictability, and duration. Nevertheless, to decide when and how policy should intervene it is essential to understand the nature of employment dynamics. In this paper we address this question by untangling the drivers of employment in Norway. *We are grateful for valuable comments from Knut Are Aastveit, Karsten Gerdrup, Ørjan Robstad, Nicolò Maffei-Faccioli and Lorenzo Mori. The views expressed in the article are solely those of the authors and cannot be attributed to Norges Bank and to the European Central Bank. †Norges Bank.: [email protected]. ‡European Central Bank.: [email protected]. §Norges Bank.: [email protected]. 1 Conceptually, it is useful for policy purposes to distinguish between trend and cyclical variation in employment. The trend component is interpreted as representing slowmoving structural factors, whereas the cycle component is associated with short-term economic fluctuations. Both components determine the current level of employment, but differs in terms of policy prescriptions. While the cycle component is primarily the concern of central banks and stabilization policies, the overall trend is potentially a matter for long-term labor market policies. Due to this division of policy, it necessary to assess in real time the relative importance of these factors. There are several ways of performing such an assessment. One approach is by detrending the aggregate time series using statistical filters such as for example the Hodrick and Prescott (1997)1method. Prior to the COVID-19 pandemic, however, Norges Bankadopteda morestructuralapproachbased onthe assumptionthat theemployment trend was driven entirely by demographic changes. In particular, the method involved fixing age-specific employment rates to their levels in a given reference year. Those levels were interpreted as reflecting their corresponding trend levels. In this setting, aggregate trend employment thus moves over time only because of changes in the agecomposition of the population. Deviations from this trend are then attributed to cyclical variations. Importantly, the reference year was deemed to be neutral with respect to the cycle, meaning that the cycle component was assumed to be zero in that year. In this paper we implement a methodology combining the two approaches. We first define demographic groups across a combination of age, sex and education levels. On the disaggregated data we then preform a trend-cycle decomposition using a Bayesian VAR in the spirit of Beveridge and Nelson (1981), allowing the trend to stochastically change over time.2Finally, we construct an aggregate trend as a size-weighted average of the group-specific trends. Thus, in contrast to the earlier Norges Bank approach, movements in the estimated aggregate trend arise from both demographic shifts and structural changes in group-specific trends.3Compared with simply de-trending aggregate employment directly, our bottom-up approach is better suited in settings characterized by large demographic changes and heterogeneous employment dynamics across popu- 1. See e.g. Veracierto (2008) for an application to the labor force participation and Krusell et al. (2017) on the employment rate. For alternative filtering methods, see e.g. Crump et al. (2019) and D’Amuri et al. (2021). 2. Our method is related to several recent contributions that allow for stochastic trends when decomposing macroeconomic data such as labor market and inflation dynamics (see e.g., Del Negro et al., 2017; Crump et al., 2019; Kamber and Wong, 2020; Ascari and Fosso, 2021; Hasenzagl et al., 2022 ). 3. An alternative which also allows for time varying trend movements is the approach adopted by the US Congressional Budget Office (Montes, 2018) based on Aaronson et al. (2006). In this approach the the employment rates of population sub-groups are estimated on age and cohort fixed effects, and various time-varying structural and cyclical co-variates. 2 lation sub-groups.4This characterization is particularly fitting for Norway, as shown in Bhuller and Eika (2020).5 We employ our method on Norwegian administrative data over the period 1984-2019, from which we identify the job status of individuals at the monthly level. Individuals are then allocated to one of 30 demographic groups, defined based on a combination of five age groups, three education levels and sex. Within each group, we aggregate the individual level data to create 30 quarterly group-specific time series for employment rates on which we perform a trend-cycle decomposition using the Bayesian VAR. Our estimated trends pick up known drivers of movements in the historical employment rate, such as changes in female labor participation and increased old-age participation. We also identify falling participation rate of low-educated workers. Moreover, the cyclical component resulting from the estimation seems to capture key business cycle movements and is highly correlated with Norges Banks’ estimate of the output gap. We contrast our method with the previous approach adopted by Norges Bank, and show that the two methods provide different assessments of both the historical evolution and current assessment of the cyclical employment gap. In recent years, the difference arises primarily as a result of our method capturing an increased trend employment among older workers.6 The remainder of the article is organized as follows. In Section 2we describe our data and how we measure employment at the individual and group level. In Section 3we outline ourstatistical model, whileSection4presents thedecompositionestimates. We conclude by discussing the policy applications in Section 5. 2 Data Data sources To create time series on employment rates disaggregated by age, education and sex we rely on Norwegian administrative data from Statistics Norway. The main data source is the employer-employee (EE) register providing us with start and end date of the near universe of wage contracts in Norway between 1984 and 2019. Individual’s wage contracts are combined with background information on education and age using unique and anonymized personal identifiers. 4. Since Perry (1971) the bottom-up approach has been commonly adopted when studying labor market dynamics. 5. Bhuller and Eika (2020) decompose the decline in aggregate employment in Norway over the period 2000-2017, and show that accounting for both demographic changes and changes within demographic groups are important for explaining the aggregate decline. 6. This old-age employment growth has in other work been linked to the work-incentivizing features of the 2011 Norwegian pension reform, see e.g. Hernæs et al. (2016) and Galaasen and Kruse (2023). 3 Definition of employed A person is defined as employed in a given month if he or she has at least one active employment contract. If no employment contract is observed in any month of a given year, we cross check using the annual tax records to assess a person’s employment status. An individual is then considered employed if (i) the annual salary income exceeds roughly USD 10,000 in 2023 nominal terms (the National Insurance basic amount) or (ii) net income above 1 basic amount from self-employment is reported that year.7 Demographic groups We split our sample into smaller units based on combinations of demographic characteristics. In particular we consider bins based on age, sex and education level. We consider five age groups, 16-24, 25-39, 40-54, 55-64 and 65-74 years of age, and three education groups, based on individuals’ highest attained education level. The high-education group corresponds to a university degree, the medium-education group to a high school diploma, and the low-education group are those with less than high-school. A person’s education characteristic is a fixed attribute, equal to the highest level of education observed for that person over the entire sample period.8. This will cause some right-censoring issues for the younger cohorts, as we typically do not know the final education status by the time our sample ends. Typically many in the youngest cohort after 2015 will be classified in the lowest education group. When estimating the model we therefore leave out the youngest cohorts during the later years in the sample to mitigate this problem. Breaks in the data Benchmarking against the official aggregate employment series revealsthatthe coverageof employmentrelations inthe EE registerchangesovertime, progressively becoming more comprehensive toward the end of our sample period. However, our use of the tax data as an additional source for identifying employed persons almost entirely eliminates the discrepancy between our aggregate series and official statistics. Still, prior to 1993 we are unable to draw on the tax data to identify employment status among those missing from the EE register. This generates some discrepancy between our aggregate series and official statistics prior to 1993, as well as a jump in the disaggregate series from December 1992 to January 1993. As our goal is to study trends in aggregate employment, we break-adjust our employment series such that our aggregate employment time series matches official statistics 7. As the detailed tax records start in 1993, prior to this year we cannot perform the last two steps. 8. For example, individuals who obtained a university degree in 2019 will be defined as highly educated over the entire sample period 4 each year. The break adjustment is performed by increasing employment in each demographic group by the same factor, thus keeping the relative shares across the groups constant. To illustrate, we have N demographic groups indexed by i with size and employment rates denoted as (popit,Eit)in year t. We thus adjust the employment share with a factor (1 + ft)such that: N X i=1 popitEit(1 + ft) = Eoff t(1) where Eoff tis aggregate number of employed from official statistics. Thus, the breakadjusted employment in year t is given by: Bt=PipopitEitft. Table 1shows the number of employees identified by each source in selected years. As we see, the break adjustment (Bt) is quite large for 1990.9This is mostly due to the fact that we do not observe detailed tax statements before 1993 which is used to crosscheck the contractual employment definition. Our adjustment implicitly assumes that the missing employed Btis distributed across the demographic groups according to their relative observed employment shares. If this assumption does not hold, our disaggregated employment series before 1993 might be biased as a result. However, the break in the series from December 1992 to January 1993 is quite similar across our demographic groups, which alleviates some of this concern. Table 1: Employees by identification method (selected years) Identified by 1990 2005 2015 Observed employment contract 1697 1975 2422 Self-employment income (tax statement) > 1 Basic amount 0 134 116 Wage income (tax statement) > 1 Basic amount 0 104 25 Break adjustment 338 29 1 Employment 2035 2242 2564 Table 1displays the identification method for employment in our dataset for selected years using the method outlined in Section 2. Numbers in thousands of persons. Employment rates Employment shares by age, sex and education levels are displayed in figure 1. Some well-known trends and facts stand out, illustrating substantial groupheterogeneity in the evolution of employment rates during the time period we look at. First, the large increase in female labor force participation is evident in the left panel, 9. Similar magnitudes are observed for all years prior to 1993 5 which shows a general increase in female employment rate from the 1980s until the early 2000s. Still, males are more likely to be employed throughout the sample period. Second, the middle panel shows that the core age groups, aged 25-54, have the highest employment rates. However, during the latter part of the sample period, the two oldest age groups experience growth in their employment rates that partially reduces the age gap. Looking at education in the right panel, we see a striking increase in the educationemployment gapovertime, drivenby asubstantial declinein theemployment rate among individuals with low education. Figure 1: Employment rates by demographic groups 3 Empirical strategy The ultimate goal of this empirical exercise is to obtain an estimate of the aggregate employment trend by exploiting the wealth of information embedded in our dataset. As mentioned above, the dataset can be described on the basis of three layers of disaggregation, from the less to the most disaggregated layer: (i) sex, (ii) age, and (iii) education. In order to get an estimate of the aggregate employment trend, we use a bottom-up approach and proceed as follows. First, we estimate the female (male) employment trends for each education level within all age groups using the empirical model presented be- 6 Figure 5: Employment gap Figure 5plots the output gap series for Norway and our employment gap series. The output gap is Norges Banks’ official estimate, while the employment gap is measured as the difference between the actual employment rate and the trend employment rate estimated using the methodology in Section 3. 4.1 Comparison with the demographic-adjusted approach The methodology developed in this paper provides a more flexible and robust trend assessment compared with the previous Norges Bank approach. The earlier methodology is essentially a demographic-adjusted approach, consisting of two steps. In the first step, the observed group specific employment rates are assumed to be on trend in a given base year. In the second step employment is assumed to evolve over time according to actual and projected population shares. This approach is restrictive as the estimate is sensitive to both the base year and the assumption of constant group trends. As an example of this, we now contrast the trend estimate for the period 2007-2019 previously adopted by Norges Bank with our estimated trend. The comparison is shown in Figure 6. In the previous approach, indicated by the red line, the employment trend is derived by fixing employment within age groups to their observed 2013 levels, and then projecting the evolution using the observed aging of the population. In summary, our estimated trend, represented by the orange line, suggest a lower trend level in 2013 but a more positive trajectory, leading to a slightly higher employment trend in the last two years before the pandemic. The source of this difference is explained in Figure 4where we decompose the change in the estimated aggregate trend estimates since 2013. The difference is primarily driven by increased trend employment among older workers, pre- 13 sumably caused by the 2011 Norwegian pension reform which stimulated labor supply among older workers (Hernæs et al., 2016).12 The downward pressure on aggregate employment stemming from an aging population is partially offset by a rise in the estimated trend employment for older workers. Figure 6: Old and New Trend Estimate Figure 6plots the old trend where employment within age groups are held fixed against our new estimated trend. 5 Policy application This section explains how the method outlined in this paper is operationalized by Norges Bank’s monetary policy department, both in analysing the current economic situation and in making forecasts. The estimated employment gap shown in Figure 5is used when assessing the current temperature of the economy. The application to forecasting is less straightforward as additional elements are needed to forecast a trend level for actual employment. The basis for the forecast is individual trends for each 30 subgroups. We apply the random walk assumption from the model in Section 3and assume that all trends remain constant from the latest observation in the data. Then we forecast the size of each group by the population forecast made by Statistics Norway. These population forecasts are not reported separately by education groups, so we add the assumption that education levels within age groups follow the trend from the last five years of reliable data (earlier than the end of the sample for younger cohorts). The forecasted 12. To account for the effect of the old age pension reform, the previous Norges Bank approach imposed a judgement-based increase in the trend employment of older workers over time. 14 group size and the estimated trend levels gives us a forecast on the level of domestic employment. For comparison with actual employment we add forecasted non-resident employment to the trend. Finally, the forecasted trend could be subject to judgement-based changes based on the assessment of other structural factors. For example, if our view on the NAIRU (non-accelerating inflation rate of unemployment) changes this could be reflected in our projected trend for employment. 6 Summary The purpose of this paper has been to document the current method adopted by Norges Bank for estimating the trend level of aggregate employment. The method consists of a bottom-up approach, whereby we first estimate disaggregated trend levels for 30 demographic sub-groups using a Bayesian VAR, and the recover the aggregate trend as the population weighted average of the disaggregated trends. Compared with simply de-trending aggregate employment directly, our bottom-up approach is better suited in settings characterized by large demographic changes and heterogeneous employment dynamics across population sub-groups. The reason is that it allows for both demographic changes and changes in trend employment within demographic groups to affect the aggregate trend level. The method outlined in this paper replaces the previous Norges Bank approach for decomposing aggregate employment dynamics into cycle and trend components. We have illustrated how this the updated method improves and changes the assessment of the historical employment dynamics, and the contemporaneous trend level, by allowing group-specific trends to be time-varying. Finally, wehaveevaluated howour estimatedcyclicalcomponent co-moveswith other measures of cyclical variation in activity. In particular, our employment gap (deviation from trend employment) shows strikingly similar dynamics since the 1990s to the more traditional output gap measure. References Aaronson, S., B. Fallick, A. Figura, J. F. Pingle, and W. L. Wascher. 2006. “The recent decline in the labor force participation rate and its implications for potential labor supply.” Brookings Papers on Economic Activity 2006 (1): 69–154. 15 Ascari, G., and L. Fosso. 2021. The inflation rate disconnect puzzle: On the international component of trend inflation and the flattening of the Phillips curve. Working Paper 733. De Nederlandsche Bank. Beveridge, S., and C. R. Nelson. 1981. “A new approach to decomposition of economic time series into permanent and transitory components with particular attention to measurement of the ‘business cycle’.” Journal of Monetary economics 7 (2): 151–174. Bhuller, M., and L. Eika. 2020. “Nedgang i sysselsettingen fra 2000–2017: –kan endringer i befolkningssammensetningen forklare alt?” Søkelys på arbeidslivet 37 (1-2): 20–37. Carter, C. K., and R. Kohn. 1994. “On Gibbs Sampling for State Space Models.” Biometrika 81 (3): 541–553. Crump, R., M. Giannoni, S. Eusepi, and A. Sahin. 2019. “A Unified Approach to Measuring u.” Brookings Papers, 143. Del Negro, M., D. Giannone, M. P. Giannoni, and A. Tambalotti. 2017. “Safety, liquidity, and the natural rate of interest.” Brookings Papers on Economic Activity 2017 (1): 235–316. Durbin, J., and S. J. Koopman. 2002. “A Simple and Efficient Simulation Smoother for State Space Time Series Analysis.” Biometrika 89 (February). D’Amuri, F., M. De Philippis, E. Guglielminetti, and S. Lo Bello. 2021. “Natural unemployment and activity rates: flow-based determinants and implications for price dynamics.” Bank of Italy Occasional Paper, no. 599. Galaasen, S. M., and H. Kruse. 2023. Spousal Spillovers in the Labor Market: A Structural Assessment. Working Paper 14. Norges Bank. Giannone, D., M. Lenza, and G. Primiceri. 2015. “Prior Selection for Vector Autoregressions.” The Review of Economics and Statistics 97 (2): 436–451. Hasenzagl, T., F. Pellegrino, L. Reichlin, and G. Ricco. 2022. “A Model of the Fed’s View on Inflation.” Review of Economics and Statistics 104 (4): 686–704. Hernæs, E., S. Markussen, J. Piggott, and K. Røed. 2016. “Pension reform and labor supply.” Journal of Public Economics 142:39–55. Hodrick, R. J., and E. C. Prescott. 1997. “Postwar US business cycles: an empirical investigation.” Journal of Money, credit, and Banking, 1–16. 16 Kamber, G., and B. Wong. 2020. “Global factors and trend inflation.” Journal of International Economics 122:103265. Krusell, P., T. Mukoyama, R. Rogerson, and A. Şahin. 2017. “Gross worker flows over the business cycle.” American Economic Review 107 (11): 3447–3476. Montes, J. 2018. CBO’s Projection of Labor Force Participation Rate. Working Paper 04. Congressional Budget Office. Perry, G. L. 1971. “Labor force structure, potential output, and productivity.” Brookings Papers on Economic Activity 1971 (3): 533–578. Veracierto, M. 2008. “On the cyclical behavior of employment, unemployment and labor force participation.” Journal of Monetary Economics 55 (6): 1143–1157. Villani,M.2009.“Steady-statepriorsforvectorautoregressions.” Journalof AppliedEconometrics 24 (4): 630–650. A Algorithm The estimation is conducted following a Bayesian approach. The Gibbs sampler is structured according to the following steps: 1. Retrieve the distribution of latent states conditional on all the other parameter of the model ¯ E0:T,ˆ E–p+1:T|vec(Φ),Σu,Σε,E1:T Draws for the latent states can be obtained by using Durbin and Koopman (2002)’s simulation smoother. In addition, we also need to draws the initial condition ¯ E0 and ˆ E–p+1:0 in order to estimate the parameters in 3and 4in the next steps. 2. Draw the parameters of vec(Φ),Σu,Σεconditional on the latent states vec(Φ),Σu,Σε|¯ E0:T,ˆ E–p+1:T,E1:T. For given ¯ E0:Tand ˆ E–p+1:T,3and 4are standard. In addition, we also already know the autoregressive matrices of the trend block in 3. Theposteriordistributionofthecovariance matrixofpermanentshocksΣuis given by p(Σu|¯ E0:T) = IW(Σu+Su, κu+T), 17 where Su=PT t=1(¯ Et–¯ Et–1)(¯ yt–¯ Et–1)0is the empirical covariance matrix of permanent shocks to the trends. The posterior distribution of the stationary parameters in vec(Φ) and Σεis given by p(Σε|ˆ E0:T) = IW(Σε+Sε, κε+T), p(vec(Φ)|Σε,ˆ E0:T) = N  vec(ˆ Φ),Σε⊗  T X t=1 ˆ xtˆ x0 t+ Ω–1  –1  , where ˆ xt= (ˆ E0 t–1,...,ˆ E0 t–p)0collects the VAR regressors, ˆ Φ =   T X t=1 ˆ xtˆ x0 t+ Ω–1   T X t=1 ˆ xtˆ E0 t+ Ω–1Φ ,Sε= T X t=1 εtε0 t+ (ˆ Φ–Φ)0Ω–1(ˆ Φ–Φ), and εt=ˆ Et–ˆ Φ0ˆ xtare the residuals of the stationary block of the model. 18