Okun's law under the demographic dynamics of the Turkish labor market

Coşar, Evren Erdoğan; Yavuz, Ayşe Arzu Article Okun's law under the demographic dynamics of the Turkish labor market Central Bank Review (CBR) Provided in Cooperation with: Central Bank of The Republic of Turkey, Ankara Suggested Citation: Coşar, Evren Erdoğan; Yavuz, Ayşe Arzu (2021) : Okun's law under the demographic dynamics of the Turkish labor market, Central Bank Review (CBR), ISSN 1303-0701, Elsevier, Amsterdam, Vol. 21, Iss. 2, pp. 59-69, https://doi.org/10.1016/j.cbrev.2021.03.002 This Version is available at: https://hdl.handle.net/10419/297936 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/ Okun’s law under the demographic dynamics of the Turkish labor market Evren Erdo gan Cos¸ar, Ays¸e Arzu Yavuz * Central Bank of the Republic of Turkey, Istiklal Caddesi No: 10, 06100, Ankara, Turkey article info Article history: Received 9 July 2020 Received in revised form 7 March 2021 Accepted 8 March 2021 Available online 18 March 2021 JEL classification: E32 J21 J64 Keywords: Okun’slaw Demographic composition Turkey abstract This study examines the asymmetric relationships between demographic characteristics of labor market variables and Gross Domestic Product (GDP) in the Turkish economy. Both expansions and recessions are considered in a Markov Switching (MS) model, using quarterly data between 1989 and 2019. Okun’s coefficients are estimated for the different age groups, genders and education levels. The results reveal that men are more likely to lose their jobs during recessions in Turkey whereas unemployment rates for 25-39 year-olds and those with at least university degrees are the least affected groups. There is also asymmetry within and between states across the demographic groups due to GDP phases. The study also investigates the gender dynamics of labor force participation rates (LFPR) as a fundamental determinant of unemployment rate. According to the MS models, LFPR responds significantly and positively to GDP expansions for men whereas it is significant and negative for women. That is, as economic activity begins to recover after a recession, Turkish women leave the labor force as secondary income earners. ©2021 Central Bank of The Republic of Turkey. Production and hosting by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). 1. Introduction Okun (1962) examines the relationship between Gross Domestic Product (GDP) and total unemployment rate for the United States of America (USA), but without acknowledging that demographic characteristics might affect this rule of thumb. In fact, Okun’s coefficient may vary by age, gender or education levels. Thus, researchers have recently conducted many studies of the changing response of labor market variables to GDP developments in terms of specific demographic groups in the labor force. Other studies have focused on the asymmetrical responses of different groups. In this study, we try to determine the cyclical asymmetry of unemployment rates for age, gender and education level in the Turkish labor force using Markov Switching (MS) regime model. The study also investigates the gender dynamics of labor force participation rates (LFPR) as a fundamental determinant of unemployment. Understanding variations across these groups may be crucial for determining the overall dynamics of the Turkish labor market in order to implement appropriate active labor market policies. This study’s main contribution is to shed light to asymmetric relationships between cyclical phases of Turkish GDP and unemployment data by age, gender and education level, using quarterly MS models for 1989e2019 period. We underline three main topics as a case study which is focusing on Turkey for the literature: demographic dynamics, size of the Okun’s coefficient and the asymmetric relationship both within and between states. One of the other specific of the paper is the investigation of the impact of the recessions and expansions on the demographic of Turkish labor market variables. The analysis at the age, gender and education level allows us to draw some policy implications and identify the vulnerable groups, which are female, less educated and young. We conclude that the output-unemployment relationship is asymmetrical for demographically different unemployment rates. That is, Okun’s coefficients are more sensitive to negative than positive output changes. More specifically, the coefficients are largest for the youngest cohort in absolute terms during recessions. While the estimated Okun’s coefficients for females are biased due to low participation rates, men are more sensitive to GDP cycles. Our findings indicate that people with more than a high school education level are least affected by transitions between different phases of GDP. To summarize, the results show that men are more likely to lose their jobs during recessions in Turkey, while workers *Corresponding author. E-mail addresses: [email protected]ov.tr (E. Erdo gan Cos¸ ar), arzu.yavuz@ tcmb.gov.tr (A. Arzu Yavuz). Peer review under responsibility of the Central Bank of the Republic of Turkey. Contents lists available at ScienceDirect Central Bank Review journal homepage: http://www.journals.elsevier.com/central-bank-review/ https://doi.org/10.1016/j.cbrev.2021.03.002 1303-0701/©2021 Central Bank of The Republic of Turkey. Production and hosting by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http:// creativecommons.org/licenses/by-nc-nd/4.0/). Central Bank Review 21 (2021) 59e69 aged 25e39 years and those with university-level qualifications or higher are the least affected by transitions between different phases of GDP. There is also within and between states asymmetry for age, gender and education groups. The MS model results display that there is a statistically significant responses of LFPR for both genders. However, its direction changes with gender. During expansions, men’s LFPR increases whereas women’s decreases. The evidence from Turkey is overall in line with the evidence from other countries. For the USA, many studies have shown deviations from full employment more strongly impact men, teenagers and lower educated people (Hyclak and Lynch, 1984;Ewing et al., 2002;Mincer, 1991). Bell and Blanchflower (2011) point out that the USA and United Kingdom (UK) unemployment rates of young people aged 16e24 have risen more severely during the recessions than other age groups. Bonaventura et al. (2018) analyse unemployment rate differences for male and female unemployment rates across the Italian regions. According to their estimation results, while in Northern regions, the change of female unemployment is more sensitive to the variation of GDP as compared to the male unemployment, the opposite occurs in the Southern regions, where the Okun coefficient is insignificant for women. Addabbo et al. (2015) investigate gender differences in the labor force participation decisions for Italy and Spain. Their results show a strong countercyclical added-worker effect for women, in contrast with a procyclical discouraged-worker effect for men. Though the added-worker effect prevails for women in Spain, in Italy the discouraged worker effect dominates. And married men show a low reaction to their partner’s labor condition with respect to women in both countries. Belaire-Franch and Peiro (2015),ina very similar study to ours, apply a MS regime model to show that male unemployment is much more sensitive to business cycle fluctuations than female unemployment, and that this sensitivity is higher in the UK than in the USA. An important point is that employment stability rises with education level because better educated workers receive more on-the-job training and accumulate more human capital. Consequently, lower-educated workers tend to be the first to lose their jobs, as demonstrated by many studies for the USA (Mukoyama and Sahin, 2006;Elsby et al., 2010; Hoynes et al., 2012;Zhang et al., 2015;Cairo and Cajner, 2017). Similar findings have been reported for OECD and Eurozone countries. Recessions have a greater effect on workers aged 15e24 or over 50 (Choudhry et al., 2012;Dietrich, 2012;Hutengs and Stadtmann, 2013;Dunsch, 2016;Dixon et al., 2017;Bruno et al., 2017;Zanin, 2014,2018), those with less education (PollmannSchult, 2005;Snieska et al., 2015;Askenazy et al., 2015) and men (Verick, 2009;Hutengs and Stadtmann, 2014;Brincikova and Darmo, 2015;Van Ours, 2015;Dunsch, 2017). From their estimations of time-varying Okun’s coefficients for South Korea for 1980e2014, Kim and Park (2019) conclude that the absolute value of the coefficient is higher for men and the 15e24 year age group than other subgroups. Based on the 1997 Asian crisis, they also report asymmetrical coefficients, being larger in recessions than expansions for all age cohorts and both genders. LFPR is a fundamental determinant of unemployment, especially for developing countries, where female LFPR is generally lower for several reasons. First, women tend to be considered secondary income earners (added worker effect) who usually look for a job only if the household’s main bread winner become unemployed. Second, structural change that has switched economic activity from agriculture to industry and services has reduced women’s participation in the labor force (discouraged worker effect). Third, cultural factors and lower female education levels push them into low-skilled jobs. Overall, it can be difficult to distinguish whether higher female unemployment during recessions is due to the discouraged worker effect or the added worker effect. Thus, most studies also investigate LFPR during downturns, especially for females. Regarding the Turkish economy, various researchers have examined the demographic composition and educational differences in unemployment rates to underline the importance of subgroup dynamics (Erdem et al., 2012;Tansel et al., 2010). Several studies have provided Okun coefficients for Turkey, including age and gender subgroups (Verick, 2009;Dietrich, 2012;Zanin, 2014; Guclu, 2018). Erdogan Cosar and Yavuz (2019) focuses on the total employment, unemployment and participation rates using MS framework together with the long and short run relationships are examined. Using Vector Autoregression (VAR) specification, Berument et al. (2006) analyse the effects of various macroeconomic policy shocks on unemployment rates for workers with different education levels, separately for men and women, for 1988e2003. They conclude that monetary policy does not affect unemployment rates in Turkey by educational level or gender whereas income and fiscal policies do. Sahin and Kizilirmak (2007) find that age, gender, education level and marital status all significantly affect the duration of workers drawing on unemployment insurance benefits in Turkey after controlling for the effects of types of workplace, job types and regions. Specifically, their estimation results indicate that older, married, male, educated and skilled workers have shorter job search durations. Bayrak and Tatli (2016) conclude that higher education schooling rates significantly and negatively effect the youth unemployment rates. Applying nonlinear dynamics and fractional integration methods to quarterly data for 1988e2013, Gil-Alana et al. (2019) find possible nonlinearities in female unemployment rates in Turkey. They also found greater persistence for the female and youth unemployment series. Regarding LFPR, Cengiz and Sahin (2014) estimate STAR models to investigate the relationship between LFPR and growth in Turkey for 2000e2011. They report an asymmetric and nonlinear relationship, such that LFPR falls during recessions but increases gradually during expansions. They also found that female LFPR falls following economic crises. Karaoglan and Okten (2015) find that there is added-worker effect for women in Turkey. That is women whose husbands involuntarily became unemployed are more likely to participate in the labor force. Besides, with a worsening of unemployment conditions, they find a slight discouraged-worker effect for women. Lastly, Signorelli et al. (2012) investigate the relationship between LFPR and financial crises for ten upper-middle-income economies, including Turkey. They conclude that financial crises significantly reduce female LFPR. In the most similar study to ours, Guclu (2018) claims that, for Turkish economy, the sensitivity of economic growth to unemployment becomes larger as age increases; and higher for men. One of the profound differences between Guclu (2018) and our paper is that he estimates a linear model using annual data 2004e2014 whereas we employ a MS non-linear model using quarterly data for 1989e2019. To the best of our knowledge, our paper is the first to consider a non-linear model to search for asymmetries between unemployment rates\LFPR and growth in Turkey while taking into account gender, age and educational level differences. This paper is organized as follows. Section 2discusses the demographic characteristics of the Turkish labor market. Section 3 presents the data sources and explains the asymmetric effects of the phase of the business cycle on age, education and gender composition. Finally, section 4concludes. 2. Turkish labor market by age, gender and education The youth unemployment rate is substantially higher than the total unemployment rate in almost every country and Turkey is no E. Erdo gan Cos¸ar and A. Arzu Yavuz Central Bank Review 21 (2021) 59e69 60 exception. Young people are twice as likely to be unemployed than the overall population where Turkey’s youth unemployment increased from 18.4 percent in 2011 to 23.1 percent in 2018, and is probably higher if the unemployed definition is extended by discouraged workers or people under the category of “neither in employment nor in education or training”(NEET). There are also high levels of informal employment among youth. Regarding gender, female youth unemployment was 25.3 percent in 2018 but only 17.6 percent for men. More specifically, female nonagricultural youth unemployment was 30.0 percent in 2018. As Fig. 1 shows, Turkey’s male and female youth unemployment rates were somewhat similar until 2008, after which the gap widened while Fig. 2 shows that female LFPR accelerated in parallel. The main driver of Turkey’s low female LFPR is the transition to a non-agricultural economy. Since many surplus female agricultural workers lack the required skills or educational qualifications for the other sectors, they either work in the informal sector or drop out of the labor force. After the 2008 global financial crisis, partly because of the added worker effect, female LFPR increased from 23.5 percent to 34.2 percent in 2018, although this is still lower than the OECD average. A recent study, Tunali et al. (2019) inspect female LFPR in Turkey using a synthetic cohort analysis and conclude that it has been on the rise lately not only due to increasing education levels but also Turkey has reached the turning point of the U-shaped pattern in female LFPR. Turkey’s unemployment rate varies widely for different education levels and university graduates have the highest LFPR. However, concerns remain as to the quality of education and whether rising education levels reduce skill mismatch in the economy. Few academic studies have investigated whether the increase in the number of universities has made economic growth more stable and created enough new jobs to reduce youth unemployment. In addition, Turkey’s educated youth find the school-to-work transition a great challenge. As Fig. 3 shows, high school graduates have the highest unemployment rates while Fig. 4 shows that the 15e24 age cohort has the highest rate by age group. Erdem and Tugcu (2012) argues that the rise in higher education graduates are one of the factors which push up the unemployment rate in Turkey in the long-run. 3. Markov Switching model estimations for Okun’s relationship by age, gender and education 3.1. Data Seasonally-adjusted quarterly labor market variables were taken from TurkStat’s Household Labor Force Survey (HLFS) while GDP data came from TurkStat. Both the HLFS and the GDP data methodology have changed several times. HLFS data were reported semi-annually between 1988 and 1999, quarterly for 2000e2004 and monthly since then. We employed seasonal factors to fill the gap for the semi-annual data. For the GDP data, TurkStat has used Fig. 1. Turkish Unemployment Rates by Gender (%, Seasonally Adjusted, Agriculture Sector Included) Note: Four quarter moving average transformation is applied to the series given at Figs. 1 and 2. Fig. 2. Turkish Labor Force Participation Rates by Gender (%, Seasonally Adjusted, Agriculture Sector Included) Note: Four quarter moving average transformation is applied to the series given at Figs. 1 and 2. Fig. 3. Turkish Unemployment Rates by Education (%, Seasonally Adjusted, Agriculture Sector Included) Note: Four quarter moving average transformation is applied to the series given at Figs. 1 and 2. E. Erdo gan Cos¸ar and A. Arzu Yavuz Central Bank Review 21 (2021) 59e69 61 ESA 2010 methodology since 2016 with a base year of 2009. The growth rates of previously announced series were used to merge the data. The variables used in the analyses cover 1989Q1 to 2019Q2. The transition from agriculture to industry and services in Turkey has created a structural change in both urbanization and labor market variables, which blur economic inferences. To get robust results, we therefore use agriculture-excluded variables in our analyses, specifically agriculture-excluded GDP and unemployment rate series by gender (total, male, female), education (“lower than high school”,“high school and equivalents”,“university or higher”) and age cohorts (15e24, 25e39 and 40þ). However, agriculture-excluded LFPR and GDP cycles are examined by gender; other LFPR classifications, age and education, reflect the total economy. Before obtaining the cyclical components of the variables, seasonal adjustment and detrending filters are applied. Seasonal adjustment is performed following Tramo/Seats methodology while the trend components of the variables are eliminated using the Hodrick-Prescott (1997) (HP) filter. This is applied to the logarithm of GDP with a smoothing parameter equal to 1600 but to the seasonally adjusted level of unemployment rate and LFPR with the same smoothing parameter. Cyclical components are defined as the difference between the seasonally-adjusted variables and the HP trends of the series. The stationarity of the cyclical components of the variables is tested by the Augmented Dickey Fuller (ADF) unit root test. Although the details are not reported here, all the cyclical components of the variables are I (0) at 1 and 5 percent significance levels. We therefore use the levels of the cyclical components in the analyses. Since we investigate asymmetric behaviour along the business cycle, the sample should be as long as possible. Therefore, while examining the relationship between unemployment rate and growth, we take 1989e2019 as the sample period as this was the largest sample for which we could extend the data backwards. 3.2. Models for unemployment rate and GDP cycles The MS model following Holmes and Silverstone (2006) that includes both the expansionary and recessionary phases of GDP and demographically categorized unemployment rates includes a random variable S t , which takes values of 0 and 1 to show the state of the economy at time t, where 0 and 1 denote above and below trend unemployment rate, respectively. The model is: where unempðdiÞ c t denotes the cyclical component of unemployment rates of gender (total, male and female), age groups (15e24, 25e39 and 40þ) or education levels (“lower than high school”, “high school and its equivalents”and “university or higher”), m indicates the state-dependent constant term, and gdp cþ t and gdp c t represent the expansionary and recessionary phases of the cyclical component of logarithmic transformed GDP, respectively. In the equation, the error term, ε t , is assumed to follow ε t i:i:d:N:ð0; s 2 ε Þ and the unobserved indicator variable S t evolves according to the following first-order MS process: PðSt¼0jSt1¼0Þ¼p PðSt¼1jSt1¼0Þ¼1p PðSt¼1jSt1¼1Þ¼q PðSt¼0jSt1¼1Þ¼1q 0<p;q<1, where pand qare the fixed transition probabilities of Fig. 4. Turkish Unemployment Rates by Age (%, Seasonally Adjusted, Agriculture Sector Included) Note: Four quarter moving average transformation is applied to the series given at Figs. 1 and 2. Table 1 Interpretation of coefficients. Expansion in GDP Recession in GDP gdp cþ t gdp c t State 0: Expansion Unemployment rate below trend b þ 0 b  0 State 1: Recession Unemployment rate above trend b þ 1 b  1 EunempðdiÞc t St¼½ð1StÞ m 0þSt m 1þð1StÞ b þ 0gdpcþ tþð1StÞ b  0gdpc tþSt b þ 1gdpcþ tþSt b  1gdpc t þð1StÞX l i¼1 d iunempðdiÞc tiþStX l i¼1 a iunempðdiÞc tiþεt (1) E. Erdo gan Cos¸ar and A. Arzu Yavuz Central Bank Review 21 (2021) 59e69 62 being in state 0 or 1. 1 The states are defined according to unemployment rate. The coefficients b þ 0 , b  0 , b þ 1 and b  1 , which represent the elasticity of unemployment rates to GDP phases in different states, are expected to take negative values (Table 1). In the regression, unemployment rate is the dependent variable and the states denote the expansions (unemployment rate below trend) and recessions (unemployment rate above trend) of the unemployment cycle. In the table, b þ 0 and b þ 1 represent the response of unemployment to the expansionary phase of GDP while b  0 and b  1 denote the response of unemployment to the recessionary phase of GDP, respectively. Table 2 summarizes the MS model estimation results for different gender unemployment rates. In general, the coefficients are statistically significant and have the expected negative sign, indicating an inverse relationship between economic growth and unemployment. For the total population, when unemployment rate is below trend, one unit increase in GDP expansion cause 0.051 unit decrease and one unit increase in GDP recession (a deeper recession) cause 0.095 unit increase in unemployment rate cycle. Similarly, when unemployment rate is above trend, one unit increase in GDP expansion cause 0.108 unit decrease and one unit increase in GDP recession (a deeper recession) cause 0.127 unit increase in unemployment rate cycle. Again for the total population, a comparison of coefficients reveals that unemployment responds more strongly to recessions than expansions in GDP while the estimation results indicate between-state asymmetry. Comparing the elasticities, b þ 1 and b  1 are higher than the other coefficients in absolute terms. That is, both male and female unemployment rates are more responsive to GDP expansion and recession phases in state 1. In Table 2 MS model estimation results by gender. Dependent variable: Cycle of Unemployment Rate (unemp c t ) State 0 ¼Expansion (Unemployment Rate Below Trend) Total Male Female Explanatory Variable Coefficient Estimate & St. Error Estimate & St. Error Estimate & St. Error Constant m 0 0.002** (0.001) 0.001 (0.001) 0.001 (0.001) gdp cþ t b þ 0 0.051** (0.023) 0.003 (0.027) 0.019 (0.039) gdp c t b  0 0.095*** (0.024) 0.002 (0.034) 0.090* (0.053) dep:var: c t1 d 0 0.89*** (0.08) 0.74*** (0.073) 0.881*** (0.073) State 1 ¼Recession (Unemployment Rate Above Trend) Total Male Female Explanatory Variable Coefficient Estimate & St. Error Estimate & St. Error Estimate & St. Error Constant m 1 0.005*** (0.002) 0.001 (0.001) 0.003 (0.004) gdp cþ t b þ 1 0.108** (0.052) 0.310*** (0.078) 0.317** (0.128) gdp c t b  1 0.127** (0.048) 0.209*** (0.057) 0.150 (0.190) dep:var: c t1 a 1 0.64*** (0.12) 0.579*** (0.134) 0.246** 0.126 Dummy variables are included in the regression. The residual diagnostics related with autocorrelation and normality are passed. *,** and *** denote statistical significance at 10, 5 and 1 percent, respectively. Number of observations are 122. Table 3 Parameter asymmetry tests by gender. Null and alternative hypotheses Type of asymmetry Wald F-stat values Total Male Female H 1 0 : b þ 0 ¼ b  0 within 0.13 1.25 1.15 H 1 A : b þ 0 s b  0 H 2 0 : b þ 1 ¼ b  1 within 1.61 0.81 0.45 H 2 A : b þ 1 s b  1 H 3 0 : b þ 0 ¼ b þ 1 between 5.50** 15.43*** 4.56** H 3 A : b þ 0 s b þ 1 H 4 0 : b  0 ¼ b  1 between 14.00*** 10.13*** 0.19 H 4 A : b  0 s b  1 (*), (**) and (***) denote statistical significance at 10, 5 and 1 percent, respectively. 1 Transition probabilities are calculated using time varying transition model and mean model. These models assume that intercept terms and the coefficients given at Table 1 are statistically different from zero. For the details of the model see Holmes and Silverstone (2006). We choose this model based on its economic meaning but not solely on its statistical significance. But for robustness issue, we test all the transition probability models given in Holmes and Silverstone (2006). Test results also lead to the use of time varying transition and mean model. E. Erdo gan Cos¸ar and A. Arzu Yavuz Central Bank Review 21 (2021) 59e69 63 addition, the higher b  1 coefficient value in absolute terms for males indicates that males in Turkey are more likely than females to lose their jobs during recessions. Although estimation sample used in this paper doesn’t cover the late Covid-19 period, estimation results are consistent with the unemployment outlook after the epidemic. Comparison of the third quarter of the year 2020 unemployment rate to the end year value of 2019 reveals that the rise in male’s unemployment rate is much higher than the rise in female’s unemployment rate. Table 3 presents the Wald coefficient equality tests to analyse state-dependent asymmetry. The first two hypotheses predict that there is no asymmetry between the expansionary and recessionary phases of GDP in states 0 and 1, respectively. The null hypothesis of the first two hypothesis (i.e. that there is within state asymmetry) could not be rejected for either gender. In other words, the expansionary and recessionary phases of GDP have the same effects on unemployment in both state 0 and in state 1. Regarding the third and fourth hypotheses, there was evidence of asymmetry between the states for total and male unemployment. That is, the effect of expansionary phase in state 0 is statistically different from the Table 4 MS model estimation results at age cohorts by gender. Dependent Variable: Cycle of Unemployment Rate According to Age Cohorts and Gender State 0 ¼Expansion (Unemployment Rate Below Trend) 15e24 25e39 40þ Explanatory Variable Coefficient Estimate &St. Error Estimate &St. Error Estimate &St. Error Male Female Male Female Male Female Constant m 0 0.003** (0.001) 0.015** (0.006) 0.006*** (0.001) 0.003** (0.001) 0.003** (0.001) 0.002*** (0.001) gdp cþ t b þ 0 0.160** (0.061) 0.427 (0.288) 0.207*** (0.061) 0.208*** (0.043) 0.060 (0.039) 0.013 (0.009) gdp c t b  0 0.295*** (0.086) 0.261** (0.100) 0.128** (0.062) 0.639*** (0.096) 0.125*** (0.032) 0.034** (0.015) dep:var: c t1 d 0 0.444*** (0.096) 0.056 (0.340) 0.239* (0.135) 0.010 (0.094) 0.472*** (0.082) 0.886*** (0.038) State 1 ¼Recession (Unemployment Rate Above Trend) 15e24 25e39 40þ Variable Coefficient Estimate &St. Error Estimate &St. Error Estimate &St. Error Male Female Male Female Male Female Constant m 1 0.010*** (0.003) 0.004* (0.003) 0.003*** (0.001) 0.002 (0.002) 0.004*** (0.001) 0.004*** (0.001) gdp cþ t b þ 1 0.724** (0.265) 0.174** (0.085) 0.207*** (0.043) 0.065 (0.075) 0.060* (0.034) 0.038** (0.016) gdp c t b  1 1.059*** (0.217) 0.631*** (0.178) 0.397*** (0.046) 0.102** (0.066) 0.212*** (0.044) 0.001 (0.051) dep:var: c t1 a 1 0.252 (0.266) 0.258 (0.174) 0.254** (0.106) 1.001*** (0.156) 0.291* (0.168) 0.685*** (0.119) Dummy variables are included in the regression. The residual diagnostics related with autocorrelation and normality are passed. *,** and *** denote statistical significance at 10, 5 and 1 percent, respectively. Number of observations are 122. Table 5 Parameter asymmetry tests for age cohorts by gender. Null and alternative hypotheses Type of asymmetry Wald F-stat values Gender 15e24 25e39 40þ H 1 0 : b þ 0 ¼ b  0 Male within 1.66 1.29 1.09 H 1 A : b þ 0 s b  0 H 1 0 : b þ 0 ¼ b  0 Female 0.25 17.59*** 6.75** H 1 A : b þ 0 s b  0 H 2 0 : b þ 1 ¼ b  1 Male within 1.70 9.14*** 8.95*** H 2 A : b þ 1 s b  1 H 2 0 : b þ 1 ¼ b  1 Female 4.20** 2.11 0.42 H 2 A : b þ 1 s b  1 H 3 0 : b þ 0 ¼ b þ 1 Male between 4.18** 0.02 0.13 H 3 A : b þ 0 s b þ 1 H 3 0 : b þ 0 ¼ b þ 1 Female 1.05 9.49*** 8.27*** H 3 A : b þ 0 s b þ 1 H 4 0 : b  0 ¼ b  1 Male between 10.79*** 13.57*** 2.38 H 4 A : b  0 s b  1 H 4 0 : b  0 ¼ b  1 Female 4.25** 19.04*** 0.43 H 4 A : b  0 s b  1 *,** and *** denote statistical significance at 10, 5 and 1 percent, respectively. E. Erdo gan Cos¸ar and A. Arzu Yavuz Central Bank Review 21 (2021) 59e69 64 effect of the expansionary phase in state 1. The same is true for the recessionary phase. In contrast, female unemployment does not respond differently to the recessionary phase in states 0 and 1. Table 4 presents Okun’s law estimation results for the three different age cohorts by gender. According to estimation results, male unemployment rate reacts to both GDP expansions and recessions with the expected negative sign both in state 0 and 1. The only exception belongs to the expansionary phase of GDP in state 0 for the 40þaged men. Besides, it is more responsive to GDP recession periods than GDP expansions for all three age groups. Age cohort 15e24 is the most responsive age group to GDP recessions and expansions, followed by 25e39 and 40þage cohorts, respectively. Female unemployment rate results is not as explicit as men’s unemployment rate results. In state 0, female unemployment rate responds to GDP expansions only at 25e39 age cohort. As a reverse of this, 25e39 is the only age cohort that doesn’t respond to expansion in state 1. Considering that the age of 25e39 is a high fertility period for women, a plausible explanation for this may be that women chose to stay out of the workforce in good times to look after their families. 40þage cohort is the least responsive age cohort for female unemployment. But there isn’t any clear-cut distinction between the 15e24 and 25e39 age cohorts in terms of their response to GDP cycles. When the results are evaluated according to gender differences it is observed that, response of male unemployment rate to economic fluctuations is slightly higher in absolute terms than the female unemployment rate in state 1. For state 0, there isn’t such a significant difference. That is when unemployment rate is above its trend, men are more likely to find a job (lose their jobs) with an expansion (recession) in GDP compared to female. Table 5 shows that there isn’t any within state asymmetry at all three age groups at state 0 but there is within state asymmetry for 25e39 and 40þage cohorts at state 1 for male unemployment rate. However, for women, there are more signs of within state asymmetry in state 1. Response of male unemployment rate to GDP expansions differ significantly according to the state only at the 15e24 age cohort. But, response of male unemployment rate to GDP recessions differ between states 0 and 1 for both 15e24 and 25e39 age cohorts. That is, male unemployment rate responds more severely to GDP recessions in state 1 than the ones in state 0. While, we observe statistically different response to GDP expansions in different states for 25e39 and 40þage cohorts for female unemployment. A similar difference occurs for GDP recessions between the two states for 15e24 and 25e39 age cohorts. Female unemployment rate for 15e24 age cohort responds more severely to GDP recessions in state 1 than state 0. Whereas the reverse is valid for the 25e39 age group. Similar to the previous findings, male unemployment rate reacts to GDP recessions more severely than expansions no matter what the education level is (See Table 6). In state 0, male unemployment rate responds to recessions at all education levels but it responds to expansions only at “high school”education group. But in state 1, it responds to both phases at “lower than high school”and “high school”education groups. At “university or higher”group, male unemployment rate respond only to recessions in both states. In sum, male unemployment rate seems to react GDP recessions more severely than expansions in both states and “lower than high school group”seems to be the most affected group from GDP fluctuations. The findings for female unemployment rate draws a blurred picture. In state 0, female unemployment rate for “lower than high school”group responds only to recessions; for “high school”group for expansions and for “university or higher”group to both expansions and recessions. But when unemployment rate is above its trend, that is state 1, being “university or higher”group as an exception, female unemployment rate responds to both phases of GDP. Similar to male, female unemployment rate belonging to the Table 6 MS model estimation results at education levels by gender. Dependent Variable: Cycle of Unemployment Rate According to Gender and Education Level State 0 ¼Expansion (Unemployment Rate Below Trend) Lower than high school High school University or higher Explanatory Variable Coefficient Estimate &St. Error Estimate &St. Error Estimate &St. Error Male Female Male Female Male Female Constant m 0 0.003** (0.001) 0.003* (0.002) 0.006*** (0.001) 0.008*** (0.002) 0.002** (0.001) 0.007*** (0.002) gdp cþ t b þ 0 0.033 (0.050) 0.008 (0.046) 0.205*** (0.063) 0.331*** (0.055) 0.029 (0.024) 0.155** (0.060) gdp c t b  0 0.311*** (0.068) 0.125** (0.058) 0.228** (0.084) 0.123 (0.107) 0.110*** (0.035) 0.358** (0.159) dep:var: c t1 d 0 0.527*** (0.083) 0.533*** (0.107) 0.290* (0.165) e0.632*** (0.081) 0.165 (0.132) State 1 ¼Recession (Unemployment Rate Above Trend) Lower than high school High school University or higher Variable Coefficient Estimate &St. Error Estimate &St. Error Estimate &St. Error Male Female Male Female Male Female Constant m 1 0.014*** (0.004) 0.005** (0.002) 0.001 (0.000) 0.012*** (0.002) 0.008*** (0.002) 0.010*** (0.002) gdp cþ t b þ 1 0.151** (0.075) 0.131** (0.054) 0.129*** (0.034) 0.340*** (0.057) 0.023 (0.137) 0.199** (0.079) gdp c t b  1 0.637** (0.316) 0.607*** (0.106) 0.395*** (0.046) 0.279** (0.112) 0.265** (0.106) 0.078 (0.145) dep:var: c t1 a 1 0.522 (0.492) 0.254 (0.192) 0.396*** (0.069) e0.984*** (0.272) 0.348*** (0.112) Dummy variables are included in the regression. The residual diagnostics related with autocorrelation and normality are passed. *,** and *** denote statistical significance at 10, 5 and 1 percent, respectively. Number of observations are 122. E. Erdo gan Cos¸ar and A. Arzu Yavuz Central Bank Review 21 (2021) 59e69 65 “lower than high school”seems mostly affected from recessions. But for higher education levels it is not possible to make a generalization. In state 1, expansions of GDP also seems to affect female unemployment rate in a positive way by creating new jobs for whom with a relatively higher education level. For both genders, when unemployment rate is above its trend, “lower than high school”group is the most adversely affected group from recessions. People in this group are more likely to lose their jobs during recessions. And among the highly educated group, GDP expansions seems to favour women compared to men in both states. However, in the “high school”group, economic fluctuations affect male and female unemployment rates similarly. According to Table 7, there is within state asymmetries in state 0 and state 1 for the lowest and highest education groups and “high school”group, respectively. Regarding female unemployment rate, there is within state asymmetries in states 1 and 0 for “lower than high school”and “high school”groups, respectively. There isn’tany difference between expansion phases across states for both genders. And finally we observe some between state asymmetries at different education groups for both genders. In other words, GDP recessions affect most severely unemployment rates in state 1 than state 0. 3.3. Models for labor force participation rate and GDP cycles To complement the analyses presented above, we also investigate the relationship between LFPR and GDP business cycle within the MS model framework. According to Fig. 5 and the cross-correlation analysis, LFPR moved countercyclically until 2001. That is, LFPR increases (decreases) as GDP decelerates (accelerates). After 2001, however, LFPR moves pro-cyclically, both overall and for males specifically, indicating that LFPR rises in parallel with faster economic growth. Female LFPR, on other hand, always moves countercyclically. Since female LFPR is very low in Turkey and females are mainly the household’s secondary income earners, their participation rate increases during recessions and decreases during expansions. Similarly, the relationship between unemployment and participation rates, both overall and for males specifically, reverses after 2001. That is, LFPR is positively correlated with unemployment before Table 7 Parameter asymmetry tests for education levels by gender. Null and alternative hypotheses Type of asymmetry Wald F-stat values Gender Lower than high school High school University or higher H 1 0 : b þ 0 ¼ b  0 Male within 9.83*** 0.07 3.20* H 1 A : b þ 0 s b  0 H 1 0 : b þ 0 ¼ b  0 Female 2.69 2.90*1.31 H 1 A : b þ 0 s b  0 H 2 0 : b þ 1 ¼ b  1 Male within 1.07 19.01*** 2.78 H 2 A : b þ 1 s b  1 H 2 0 : b þ 1 ¼ b  1 Female 15.41*** 0.19 2.49 H 2 A : b þ 1 s b  1 H 3 0 : b þ 0 ¼ b þ 1 Male between 0.18 1.25 0.15 H 3 A : b þ 0 s b þ 1 H 3 0 : b þ 0 ¼ b þ 1 Female 2.53 0.02 0.28 H 3 A : b þ 0 s b þ 1 H 4 0 : b  0 ¼ b  1 Male between 1.04 3.31*2.15 H 4 A : b  0 s b  1 H 4 0 : b  0 ¼ b  1 Female 16.46*** 1.46 3.26* H 4 A : b  0 s b  1 *,** and *** denote statistical significance at 10, 5 and 1 percent, respectively. Fig. 5. Cycles of unemployment rate, labor force participation rate and GDP E. Erdo gan Cos¸ar and A. Arzu Yavuz Central Bank Review 21 (2021) 59e69 66