The effect of shocks to labour market flows on unemployment and participation rates

Abstract

This article presents an analysis of labour market dynamics, in particular of flows in the labour market and how they interact and affect the evolution of unemployment rates and participation rates, the two main indicators of labour market performance. Our analysis has two special features. First, apart from the two labour market states – employment and unemployment – we consider a third state – out of the labour force. Second, we study net rather than gross flows, where net refers to the balance of flows between any two labour market states. Distinguishing a third state is important because the labour market flows to and from that state are quantitatively important. Focusing on net flows simplifies the complexity of interactions between the flows and allows us to perform a dynamic analysis in a structural vector-autoregression framework. We find that a shock to the net flow from unemployment to employment drives the unemployment rate and the participation rate in opposite directions while a shock to the net flow from not in the labour force to unemployment drives the rates in the same direction.

Keywords:

• net labour market flows
• unemployment rate
• participation rate

JEL Classification:

• E17
• E24
• J21
• J64

Acknowledgement

We thank an anonymous referee for helpful comments on a previous version of the article.

Notes

¹ Early models tended to focus solely on flows between employment and unemployment. However, in recent years, some researchers have begun to explicitly include a third state – inactive or not in the labour force in their studies. Examples include Elsby et al. (2009), Petrongolo and Pissarides (2008), Shimer (2012), Smith (2011) and Sutton (2013).

² The availability of data on gross flows between states allows researchers to examine the balance of flows between labour market states. However, this is rarely done. One of very few attempts to study net flows is that by the European Commission (2009) which examines the size of the net flows in the EU in order to assess the relative size of ‘good transitions’ (flows into employment) compared with ‘bad transitions’ (flows out of employment).

³ Recent papers on gross flows include Elsby et al. (2009), Elsby et al. (2013), Fujita and Ramey (2009), Petrongolo and Pissarides (2008) and Shimer (2012). The main purpose of these papers is to assess the relative contribution of fluctuations in unemployment inflow and outflow rates to unemployment variation. Our approach differs in that not only are we viewing labour market dynamics (including unemployment fluctuations) using net rather than gross flows but also in that we are interested in the impacts of shocks to the (net) flows on the participation rate and the employment population ratio in addition to the unemployment rate.

⁴ Schettkat (1996) provides a neat discussion of this.

⁵ This subsection draws upon some ideas in Dixon et al. (2011).

⁶ The data we use are based on matching records (responses to survey questions by the same respondents) over two successive periods, and hence by construction the population at the beginning and end of the period must be the same. This assumption is relaxed in the empirical section because our data set over time is adjusted to be stock-consistent which includes the effect of population changes.

⁷ In principle, the equilibrium condition may also be satisfied if all net flows are zero or negative (and equal). We focus on the case where the net flows are positive for two reasons. First, as we shall see in the next section, this is actually the case in the data we consider. Second, if all the net flows are negative (and equal), this would suggest that the dominant source of increases in employment is the (direct) flow from N to E. Empirically it would seem to be the case that the dominant net flow into E in any period is from U.

⁸ See, for example, Mincer (1966), Lundberg (1985) and Stephens, Jr. (2002).

⁹ Or directly becoming employed without moving for any measured time in the state of being unemployed.

¹⁰ An example might be a shock to employment due to an expansion in government spending along the lines discussed in Monacelli et al. (2010) and Brückner and Pappa (2012).

¹¹ The reasons why the ‘population represented by the matched records’ is less than 100% of the total civilian population aged 15 years and over are explored in some detail in Dixon et al. (2002).

¹² The raw data for stocks were taken from the ABS Labour force published ‘stock’ data (original, not seasonally adjusted). Seasonal adjustment was performed after the flows were made stock consistent using EViews v.7 (Irvine, CA, USA).

¹³ For more information on ‘raking’, see Abowd and Zellner (1985) and Frazis et al. (2005). See the Appendix for details on the raking procedure used.

¹⁴ The flows even when expressed relative to a stock are very noisy. For this reason, the measures using grouped data may better reveal the underlying relationships.

¹⁵ Tables of gross flows reported for the US and the UK show that net flows are in the same direction as that reported here. See, for example, studies of US data by Blanchard et al. (1990) and Boon et al. (2008), who report average flows for the years 1968–1986 and 1990–2006, respectively. For the UK, see, for example, Bell and Smith (2002) and Gomes (2012), who report average flows for the years 1996–2001 and 1996–2010, respectively.

¹⁶ The ADF statistics are $\underset{\_}{nu}$ (−18.3), $\underset{\_}{ue}$ (−6.8) and $\underset{\_}{en}$ (−27.9). The 5% critical value is −1.9, and so the null of nonstationarity is rejected.

¹⁷ Sims (1980) advocated the use of the VAR approach to model linear interdependencies among multiple time series as a way of capturing dynamic relationships without imposing strong structure. The SVAR approach allows for contemporaneous relationships, a feature absent in the traditional VAR.

¹⁸ We tested an alternative specification where the SVAR contained a mixture of net flows and change in rates rather than lagged net flows. The likelihood of that model is less than the one presented here.

¹⁹ The Ljung-Box Q-statistics for first-order autocorrelation (p-values in parenthesis) are 0.611 (0.434), 1.010 (0.315) and 0.630 (0.427).

²⁰ Of course, an increase in the youth minimum wage also has an effect on labour demand. See Boeri and van Ours (2013) for an overview of youth minimum wage studies.

²¹ Since re-interviews do not take place in Australia, we are unable to use that information to adjust for classification errors. Hence, the only adjustment we can make is for ‘margin error’. For more information on the margin errors and ‘iterative raking’, see Abowd and Zellner (1985) and Frazis et al. (2005).

²² Frazis et al. (2005) provide a neat discussion of the procedure as applied by the BLS (especially page 6).

²³ We used 10 ‘complete’ iterations which in fact went well beyond the requirement stated here for most months.

²⁴ Table 2 of Frazis et al. (2005) shows that these changes are also present in the adjusted data for the US.