For the latest Coronavirus (COVID-19) Victoria updates, restrictions, travel permits and exposure sites

A flow decomposition of the unemployment rate in Victoria

Victoria's Economic Bulletin shows the importance of labour flows between employment and unemployment in shaping the unemployment rate in Victoria over the past two decades.

Published: June 2021

Written by Omid Mousavi, Maryam Nasiri, Jiayi Wang and Bedika L Mala.[1]

Document download

Contents

Overview
1. Literature Review 
2. Data
3. Methodology
4. A Cyclical Decomposition of the Unemployment Rate in Terms of Labour Market Flows
5. Accounting for the Labour Market Heterogeneity
6. Conclusion
References

Abstract

The unemployment rate is shaped by labour flows between employment and unemployment as well as those in and out of the labour market. In this article, we use a flow-based approach to show the importance of each of these flows in shaping the unemployment rate in Victoria over the past two decades. We find that flows between employment and unemployment primarily explain the fluctuation in the unemployment rate in Victoria. We also find that flows out of the labour market can explain a larger fraction of the unemployment rate during episodes when the unemployment rate is high.

Understanding the factors (flows) that drive changes in the unemployment rate at different points in the business cycle can be important for informing labour market policy. This has been particularly evident during the coronavirus (COVID-19) pandemic. While the adverse impacts of the coronavirus pandemic on the labour market are still resolving, the policy responses such as JobKeeper and JobSeeker have focused on maintaining job matches and directly affecting the labour market flows. This may suggest future research examining responses of different labour market flows to these policies and their impact on the labour market.

Overview

The unemployment rate is a summary of different movements (flows) in the labour market that occur in a specified time period. At any point in time, a large number of Victorians move from one job to another, from employment to unemployment and vice versa. There is also a large number of people moving in and out of the labour force at any point in time. This article provides a cyclical perspective on the unemployment rate disaggregated in terms of these labour market flows. We use a flow-based approach to investigate the importance of different labour market flows in explaining the unemployment rate in Victoria over the past two decades.

To inform policy development, it is important to know the source of changes in the unemployment rate. For instance, when an increase in the unemployment rate is driven by a decrease in the number of unemployed workers finding a job then policies can be specifically targeted at improving the job-finding prospects of unemployed workers, such as job training or job search assistance programs.[2] On the other hand, when changes in the unemployment rate are mostly driven by changes in labour market participation, policies may focus on encouraging workers back into the labour market. For instance, via job creation when an increase in the unemployment rate is due to discouraged workers leaving the labour market.

We find that the unemployment rate is primarily driven by ‘turnover’, which is defined by the change in the transition of workers between employment and unemployment. This indicates that the majority of cyclical changes in the unemployment rate can be explained by changes in the job-finding rate of unemployed individuals and job separation rate of the employed individuals.

In general, entry to and exit from the labour market do not contribute to fluctuations in the unemployment rate as significantly as turnover. However, the role of labour market participation becomes more important during episodes when the unemployment rate is relatively high, with a lower probability of exiting the labour market observed in the high unemployment rate episode of late 2014.[3] Intuitively, during episodes when the unemployment rate is high, the composition of unemployment may change toward workers who have a lower tendency to exit.

The remainder of this paper is as follows. In Section 1, we review the literature. In Section 2, we explain the data. In Section 3, we outline the methodology for decomposing the unemployment rate into different labour market flows. In Section 4, we present the results of the decomposition. Section 5 describes the role of labour market heterogeneity in shaping the unemployment rate in Victoria and Section 6 concludes.

1. Literature Review 

Our work focuses on how changes in different labour market flows contribute to shaping the unemployment rate in Victoria. To the best of our knowledge, only a few studies have examined labour market flows in Australia. Our work contributes to the existing literature by using a model that, unlike previous studies, accounts for short run deviations (transitional dynamics) in the unemployment rate from its long-run trend.

A substantial body of research on labour market fluctuations focuses on turnover in the labour market – that is, transitions between employment and unemployment and vice versa. This strand of literature assumes that changes in labour market participation – that is, movements in and out of the labour force, play little to no role in driving fluctuations in the unemployment rate (Aaronson et al., 2010).[4]

Under the assumption that workers neither enter nor exit the labour force, the job-finding rate is found to have a more significant impact on the variation in the unemployment rate than the job loss rate (Shimer, 2012; Elsby et al., 2009; Mazumder, 2007). In particular, movements in the job-finding rate have been found to account for most of the variation in the unemployment rate during the last two decades (Shimer, 2012). However, the amplitude of fluctuations in the flow out of employment is larger than that of the flow into employment, implying a much larger amplitude of the underlying fluctuations in job destruction than that of job creation (Davis and Haltiwanger, 1990).

While most research has neglected transitions relating to labour force participation, Elsby et al. (2015) argues that flows relating to participation account for around one-third of the cyclical variation in the unemployment rate. Similarly, Krusell et al. (2011) found that the participation rate, employment rate and unemployment rate jointly determine the variation in the unemployment rate and suggest that a comprehensive model of the aggregate labour market should explicitly incorporate all three labour market states. This is further explored by Barnichon et al. (2012), who developed a forecasting model for the unemployment rate based on labour market flows between all three labour market states. The model produced more accurate forecasts and performed especially well during large recessions and cyclical turning points.

Among studies that examine various components of unemployment in Australia, Ponomareva and Sheen (2009) use an equilibrium model with four labour market states and estimate the associated transition flows in Australia.[5] They find that transitions within the labour market (specifically the transition from unemployment to employment) contribute more significantly to explaining variation in the unemployment rate than other transitions. Chindamo (2010) also estimates transition probabilities and finds that the decline in the job-finding rate contributed up to 10 percentage points to the economic downturns in the early 1980s and early 1990s.

2. Data

We use gross flows data that is obtained from the Labour Force Survey of the Australian Bureau of Statistics (ABS). The gross flows data provides the counts of individuals transitioning between employment, unemployment and out of the labour force by their geographic location (state), age and gender.[6]

Using the gross flows data, it is straightforward to estimate the transition probabilities associated with each flow. This is accomplished by expressing the number of people who transition from one state (e.g. unemployment) into another state (e.g. employment) as a fraction of the number of people in the original state in the previous period (unemployment in this example). More formally, the transition probabilities for each time t are estimated according to ptij=ijt/it-1 for i and j ∈ {E,U,N}.

These measures are informative about labour market dynamics and have been widely used in research on labour market dynamics. We disaggregate the flows by gender and age groups of individuals for each state, over time. Table 1 shows the sample averages for the three labour market states for the whole sample and across different genders.

Table 1 shows that, on average, an unemployed worker in Victoria has a 21 per cent probability of finding a job (PUE). However, an unemployed worker is slightly more likely to drop out of the labour market in a given month (PUN) than finding a job (23 per cent). Additionally, individuals that are not participating in the labour force are more likely to find a job (PNE) than becoming unemployed (PNU) when they begin labour market activity (i.e. searching for a job).

Further, Table 1 shows that men are more likely to find a job than women regardless of their labour market status in the previous month, i.e. they have a higher PUE and  PNE than women. Additionally, men are around 7 percentage points less likely to leave the labour market than women. This can be seen by comparing the PEN+PUN between men and women.

Table 1. Transition probabilities by gender

Transition probability

Whole sample          

Men                       

Women                       

PEE

96.36

96.89

95.72

PEU

0.90

0.97

0.81

PEN

2.74

2.14

3.47

PUE

21.30

21.96

20.54

PUU

56.15

58.11

53.94

PUN

22.55

19.93

25.52

PNE

4.61

5.03

4.34

PNU

2.84

3.35

2.51

PNN

92.56

91.61

93.15

Note: Average monthly transitions in per cent for the December 1999–March 2020.

We delve deeper into the gross flow data from the ABS and construct similar measures as shown in Table 1 for individuals in different age groups in Victoria. Figure 1 shows labour market turnover measured by PEU+PUE, labour market exit measured by PEN+PUN and labour market entry measured by PNE+PNU for different age groups of workers in Victoria. The left panel in Figure 1 shows that labour market turnover is highest among workers in the 25–34 years age group, whereas workers in the 55–64 years age group have the lowest turnover among workers in Victoria.

The right panel in Figure 1 reveals that younger workers have the highest tendency to exit the labour market. Workers aged between 25–54 years have a fairly similar tendency of exiting the labour market, while this increases among workers in the 55–64 years age group, potentially reflecting the higher incidence of retirement. Looking into the bottom panel in Figure 1, younger workers have the highest tendency of entering the labour market. Figure 1 also shows that entering the labour market decreases with age.

In general, understanding the impact of age and gender heterogeneities on the unemployment rate is not straightforward. In Section 6, we explore the role of these heterogeneities in explaining the changes in the unemployment rate by asking how the unemployment rate (counterfactual) would have changed if the labour market was formed by each of these groups.

Figure 1. Transition probabilities by age

3. Methodology

This research closely follows Shimer (2012), Elsby et.al (2015) and Elsby et.al (2020) who provide a methodological framework for decomposing the responses of the unemployment rate (along with other labour market outcomes) to different labour market flows. In particular, we are guided by Elsby et.al (2020) who also incorporate transitional dynamics over time. This specifically allows us to address the short run deviation from the long run trend behaviour of the unemployment rate. This is important given the slower return of a short run deviation of the unemployment rate to the long run trend in the data.[7]

Let us begin with a simple discrete time Markov chain that represents the stock and flows in the labour market. This can be shown by:

OMequation1

 

     (1)

Where E, U and N denote the stocks of employed, unemployed and non-participating individuals respectively. Without loss of generality, we normalise the size of the population to one, i.e. et+ut+nt=1 in which e, u and n denote the respective shares of employment, unemployment and non-participation in the population. Specifically, we assess the response of the unemployment rate to changes in (i) labour market turnover (measured by PUE and PEU); (ii) labour market exit (measured by PUN and PEN) and (iii) labour market entry (measured by PNE and PNU).

To simplify the notation, denote:

OMequation1a

 


and 

OMequation1b

 


Using this we can simplify Equation (1) to:

OMequation2

       (2)

 

Or 

OMequation2a

 

We can also simplify the notation further by defining:

OMequation2b

 

where I is an identity matrix. In doing so, Equation (2) can be written as:

OMequation3

      (3)

Using Equation (3) and for fixed elements of the transition matrix Pt, we can show that the steady state elements of vector is:

OMequation3a

 

In other words, the elements of the state vector converge to the flow steady state (long run equilibrium) implied by this equation. Elsby et.al (2020) show that Equation (3) can be written as following[8]:

OMequation4

    (4)

It can be shown that the transitional dynamics, which is a change in the state from the steady state (long run trend), can be written as:

OMequation5

     (5)

Which is reflected in Equation (4). This implies that current change in the state (i.e. St) depends on the past changes in the deviation of the state from its steady state. In addition, it can be shown that the second term in Equation (4) accounts for changes in the steady state over time as a function of different transition probabilities.

It should be emphasized that Equation (4) is additive in terms of changes in the transitional probabilities. This result helps us to measure the contribution of each transition probability to changes in the unemployment rate, which we examine in the next section. Specifically, we will measure how changes in the elements of the transition probability matrix affects changes in the states (e.g. unemployment).

4. A cyclical decomposition of the unemployment Rate in terms of labour market flows

Our sample covers the period of December 1999 to March 2020. Although this time period does not include a reported recession in Australia, we can observe at least three significant episodes during which the unemployment rate increased significantly in Victoria.

The first episode is associated with the ‘dot com’ bust. During this episode, the unemployment rate reached around 7 per cent in the early 2000s.[9] The second episode is associated with the GFC, when the official unemployment rate reached about 6 per cent in August 2009 from a low of 4.2 per cent in the previous year.[10]

The most recent episode in our sample occurs in late 2014, where the unemployment rate reached around 7 per cent in October 2014. Although the peak in the unemployment rate was not associated with a recession, the peak in the unemployment rate was followed by a temporary increase in labour market participation. In addition, this episode was not associated with any noticeable change in the number of job advertisements.[11] This suggests that changes in the flows into and out of the labour market (i.e. entry and exit to the labour market) were potentially the primary sources of the increase in the unemployment rate over this time period.

In the remainder of this section, we analyse how different labour market flows explain changes in the unemployment rate during each episode.

We begin our analysis by showing how the unemployment rate delivered by the model tracks data over time. Figure 2 compares the unemployment rate derived from the model with data. It shows that the model performs well in approximating the unemployment rate in Victoria (correlation ≈ 0.97). Therefore, we can confidently use the model to decompose the unemployment rate in terms of different labour market flows.

Figure 2. Unemployment rate in Victoria - total decomposition

We proceed with highlighting the importance of different transition probabilities in shaping the unemployment rate. In each decomposition, we only allow the associated transition probabilities to vary over time while holding the other transition probabilities fixed at their long run averages. These results are provided in Figure 3.

The decomposition in Figure 3 shows that changes in labour market turnover (i.e. the probability of finding and/or losing a job) has driven most of the variation in the unemployment rate in Victoria over the past two decades. In general, holding other transition probabilities unchanged at their long run averages, the unemployment rate is expected to increase with an increase in the probability of losing a job (PEU) and/or with a decrease in the probability of finding a job (PUE). Over business cycles, these two probabilities vary, which creates variation in the unemployment rate. Overall, the direction of change in the unemployment rate depends on the relative strength of the change in these two probabilities.

To have a better understanding of this, Panel (a) in Figure 4 shows the changes in PEU and PUE over time in Victoria. We can see that over the period leading to an increase in the unemployment rate in the late 2009’s, we observe an increase in PEU and a decrease in PUE. However, during the third episode of high unemployment rate in late 2014, although we observe an increase in the probability of losing a job for an employed person, we also observe a slight increase in the probability of finding a job for an unemployed person, which to some extent dampens the increase in the unemployment rate.

Figure 3. Unemployment rate in Victoria - transition decomposition

Figure 3 also highlights the role of labour market entry and exit in shaping the unemployment rate. In general, relative to turnover, entry and exit play minor roles in shaping the unemployment rate in Victoria over time. However, we find that exits from the labour market have a larger bearing on the unemployment rate during periods when the unemployment rate is high, such as in the early 2000s and late 2014. In general, during episodes when the unemployment rate is high, with an increase in the flow of workers out of employment toward unemployment, the composition of the unemployment pool changes toward workers who have a lower tendency to exit. This leads to a decline in the transition probability of workers from unemployment to out of the labour market (i.e. PUN) that translates into an increase in the unemployment rate. On the other hand, during times of recovery in the labour market, the composition of unemployment shifts toward workers who are more likely to exit, which results in an improvement in the unemployment rate.

The decrease in exits from the labour market when the unemployment rate is high and the increase in exits during the recovery phase can be seen from Panel (b) of Figure 4. The figure shows that in the periods leading to the peak in the unemployment rate in late 2014, the PUN declines, while it increases afterwards. In the next section, we elaborate on this point by showing the role of labour market heterogeneity in driving the unemployment rate over time.

Finally, we find a smaller role for labour market entry in explaining changes in the unemployment rate. Intuitively, a recovery in the labour market encourages more individuals to enter into the labour market. This implies that during episodes when the unemployment rate is low, we should expect to observe an increase in both PNE and PNU. On the other hand, in a slack labour market, workers may delay their entry into the labour market. Panel (c) in Figure 4 shows the changes in the probability of entry into the labour market over time. Although we can find some periods in which this hypothesis holds (such as an increase in PNE during the recovery after the peak in the unemployment rate in the early 2000s), it is difficult to propose this hypothesis to other periods. Overall, Figure 3 shows that changes in the entry into the labour market (i.e. PNE and PNU), holding other transitions unchanged, does not explain changes in the unemployment rate as significantly as the other two forces.

The findings in this section can inform policy development, as it is important to identify the economic mechanisms behind an increase in the unemployment rate. For instance, if changes in the unemployment rate are driven by workers having difficulty in finding a job, then the policy may emphasize programs that aim to help workers to search more effectively for a job such as job training or job search assistance programs. However, if the changes in the unemployment rate are primarily driven by a lower participation rate, the policy may focus on increasing the return to job search, for instance, via programs that help with job creation. 

Figure 4. Transition probabilities in Victoria

(a) Labour market turnover

(b) Labour market exit

(c) Labour market entry

Note: All series are seasonally adjusted monthly data and smoothed with a 6-month cantered moving average.

5. Accounting for the labour market heterogeneity

We proceed with analysing the role of labour market heterogeneity in explaining the fluctuation in the unemployment rate in Victoria over the past two decades. In doing so, we present the unemployment rate implied by different genders and age groups. The results provide a series of counterfactual unemployment rates where the sample is represented by a specific gender or age group.

5.1 Role of gender

Figure 5 shows the unemployment rates implied by the model when each gender group represents the entire pool. In other words, it represents the unemployment rate for the situation where only males or females participate in labour market activities.

On average, we observe a lower unemployment rate when the unemployment pool is only represented by men. In contrast, when the unemployment pool is only represented by women, the unemployment rate tends to be higher. This can be explained by a combination of (i) the higher chance of finding a job (PUE) (ii) higher probability of remaining labour market (PUN and PEN) and (iii) higher intensity of entering into the labour market (PNU and PNE) that was presented in Table 1.

A closer look at Figure 5 reveals that the variation in the unemployment rate when only men form the sample is larger than the variation in the unemployment rate when only women form the sample. This suggests that fluctuations in the unemployment rate are mainly driven by the male component. This is particularly evident during the GFC in which the response among men was much higher than women (3 vs <1 percentage points).

Figure 5. Unemployment rate in Victoria - gender decomposition

5.2 Role of age

We proceed with analysing how different age groups contribute to the changes in the unemployment rate in Victoria. As before, we compute a series of counterfactual unemployment rates in which only the relevant group constitutes the labour market, i.e. what would the unemployment rate be if all workers behaved as workers in the 34–45 years category?

Figure 6 shows that among five different age groups, workers in the 25–34 and 35–44 years age group are the ones that most closely replicate the total unemployment rate in Victoria.[12][13] Assuming a labour market where individuals characteristics are similar to workers in the 45–54 years age group, we would have expected a higher unemployment rate, whereas if the characteristics are similar to the 55–64 years age group, the unemployment rate would have been lower.

This result reflects the differential impact of entry and exit into the labour market for the different age groups. In general, older workers are less attached to the labour market (i.e. exit more) and also enter the labour market at a lower rate. Whereas, gains from having a job and forming an employment relationship is higher among younger workers so that they are more likely to stay attached to the labour market and enter at a higher rate. The role of labour market participation is particularly evident from the different unemployment rates associated with workers in the age groups of 55–64 and 45–54. The oldest group exit more and enter less relative to the total which explains lower implied unemployment rate. However, workers in the 45–54 age group exit less while entering more, which explains the high unemployment rate relative to the other age groups.

Figure 6. Unemployment rate in Victoria - age decomposition

6. Conclusion

In this paper, we ask how different labour market transition probabilities have shaped the unemployment rate in Victoria over the past two decades. We answer this question by using a labour market flow approach to decompose the unemployment rate into different labour market transitions.

We show that variation in transitions associated with losing and finding a job can significantly explain variation in the unemployment rate over time. Moreover, we show that during the specified episodes of high unemployment, the composition of the unemployment pool changes towards workers with higher incentives to stay attached to the labour market for longer periods of time (such as younger workers). This explains our finding that changes in exits from the labour market can explain changes in the unemployment rate during episodes when the unemployment rate is high.

These findings have important policy implications. In a labour market where changes in the unemployment rate are primarily driven by workers having difficulties in finding and accessing a job, the relevant policies may include ones that help workers to search more effectively for a job (such as job training). However, if the change in the unemployment rate is related to lower labour market participation, policies that focus on maintaining attachment of workers to the labour market, such as policies that focus on job creation, may be more effective. 

We also highlight the role of labour market heterogeneity in shaping the unemployment rate in Victoria. We show that when the unemployment pool is represented by men and/or workers in the age group of 25-44, the implied unemployment rate more closely follows the actual unemployment rate in Victoria.

At the time of writing this article, the adverse impacts of the coronavirus pandemic on the labour market is still resolving. Many workers have lost their jobs and many others have dropped out of the labour market. While the coronavirus (COVID-19) pandemic shock has directly (adversely) affected the transition probabilities between labour market states, the policy responses such as JobKeeper and JobSeeker have focused on maintaining job matches and maintaining the labour market participation. This may suggest future research on examining the role of these policies on the different labour market transitions and their impact upon labour market during the pandemic.

References

Aaronson, D., Mazumder, B., & Schechter, S. (2010). What is behind the rise in long-term unemployment? Economic Perspectives, 34(2).

Barnichon, R., Nekarda, C. J., Hatzius, J., Stehn, S. J., & Petrongolo, B. (2012). The Ins and Outs of Forecasting Unemployment: Using Labor Force Flows to Forecast the Labor Market [with Comments and Discussion]. Brookings Papers on Economic Activity, 83-131.

Chindamo, P. (2010), Australian Labour Market Flows Over the Business Cycle, Australian Bulletin of Labour, 127-137.

Davis, S. J., & Haltiwanger, J. (1990). Gross job creation and destruction: Microeconomic evidence and macroeconomic implications. NBER macroeconomics annual, 5, 123-168

Elsby, M. W., Michaels, R., & Solon, G. (2009). The ins and outs of cyclical unemployment. American Economic Journal: Macroeconomics, 1(1), 84-110.

Elsby, M. W., Hobijn, B., & Şahin, A. (2015). On the importance of the participation margin for labor market fluctuations. Journal of Monetary Economics, 72, 64-82.

Krusell, P., Mukoyama, T., Rogerson, R., & Şahin, A. (2011). A three-state model of worker flows in general equilibrium. Journal of Economic Theory, 146(3), 1107-1133.

Mazumder, B. (2007). New evidence on labor market dynamics over the business cycle. Economic Perspectives, 31(1).

Mortensen, D. T., & Pissarides, C. A. (1994). Job creation and job destruction in the theory of unemployment. The review of economic studies, 61(3), 397-415.

Ponomareva, N. & Sheen, J., 2010. Cyclical flows in Australian labour markets. Economic Record86, pp.35-48.

Shimer, R. (2012). Reassessing the ins and outs of unemployment. Review of Economic Dynamics, 15(2), 127-148.

Footnotes 

[1] The authors would like to thank James Brugler and Gillian Thornton for their comments. The views expressed in this paper are those of the authors and do not necessarily reflect the views of DTF.

[2] For example, via reduced job search effort or enhancement of skills required for finding a job.

[3] We investigate three episodes when the unemployment rate in Victoria is relatively high: The Global Financial Crisis (GFC), where the unemployment rate increased to about 6 per cent from a low of about 4 per cent in a short period of time, and two periods in October 2001 and October 2014 where the unemployment rate reached about 7 per cent.  

[4] These studies are informed by Mortensen and Pissarides (1994) labour search and matching model, which predominantly focuses on transitions between employment and unemployment.

[5] In particular, they model full‐time and part‐time as two separate states for employment.

[6] The Labour Force Survey contains a panel of eight rotating groups that are surveyed for eight consecutive months. It includes labour force activity data of around 52,000 people in 26,000 households. A new rotation group is introduced each month to replace an outgoing rotation group, generally from the same geographic area.  

[7] In general, following a short run deviation, the unemployment rate returns to the long run trend relatively faster in a labour market with a high turnover rate, for instance, the labour market in the United States. This implies that in a labour market with high turnover, the unemployment rate can be more closely approximated by the steady state (or long run) unemployment rate. Therefore, given the slow return in our data, we need a model that accounts for the transitional dynamics.

[8] Please see Elsby et.al (2020) for a more detailed derivation of this result.

[9] Specifically, the seasonally adjusted unemployment rate reached 7.1 per cent in October 2001. Source: ABS.

[10] Source: ABS.

[11] Source: Labour Market Portal. Please see the Labour Market Information Portal.

[12] These five age groups are: 15-24, 25-34, 35-44, 45-54 and 55-64 year old.

[13] This could also reflect the distribution effect in that this group more closely represent the distribution of unemployment in Victoria. 

Reviewed 22/06/2021
Was this page helpful?