Informal work along the business cycle: evidence from Argentina

The worker flows are obtained by exploiting the rotating-panel structure of the Argentine survey data. Individuals in a given survey are linked longitudinally to their response in the subsequent survey. This property allows us to estimate worker flows and their associated transition probabilities. For example, the probability that an unemployed worker finds a formal job can be computed as the share of the unemployed in a given survey who report in the subsequent survey that they are in formal employment. We can then compute transition probabilities between employment in formal and informal jobs, unemployment, and nonparticipation.

To build worker flows, we adopt a Markovian representation. The mapping between labor force stocks and flows is modeled as a simple discrete-time Markov chain:

(1)Xt=ℓtXt−1$$X_{t}=\ell_{t} X_{t-1}$$

where X_t =(F_t, I_t, U_t, N_t ) denotes the four labor market stocks, namely, F formal employment, I informal employment, U unemployment, and N non-participation. l_t is a square matrix of size 4, whose elements l_i,_j capture the transition probability from labor status i to labor status j.

We adjust the data along three dimensions as mentioned in Elsby et al. (2015). We first seasonally adjust gross flows using the X12 ARIMA process. We then compute transition probabilities that are consistent with the observed changes in stocks (F_t,I_t,U_t,N_t) (correction for margin error, see Appendix A.1). Finally, as gross flows provide transition probabilities observed at discrete points in time, we correct these measures for possible transitions occurring between consecutive surveys (Shimer, 2012, see Appendix A.2). We then get instantaneous transition rates (also referred to as hazard rates) λ_ij, derived from a continuous-time model of labor market transitions. Adjusted transition probabilities are then derived from instantaneous transition rates as 1 − e^{− ij}. Further details are provided in Appendix A. Time-series of transitions probabilities are displayed in Appendix B.

In this section, we lay out several preliminary facts about informal workers. The average characteristics of the sample period are presented in Table 1. Over the sample period, informality is pervasive: 37% of Argentine employees are in informal employment. Fifty-seven percent of formal workers are male versus 50% in informal employment. Women are more likely to have informal jobs than men. In addition, 8% of formal workers are aged 16–24 versus 27% in informal jobs. The incidence of informality among low-skilled workers is higher than among high-skilled workers. This result is consistent with Albertini and Terriau (2019) who use the schooling degree as a proxy for skill and La Porta and Shleifer (2014) who document the productivity gap between formal and informal firms. It also suggests that informal jobs may often be the only opportunity available to unskilled workers. Table 1 suggests that, with respect to formal workers, workers in informal jobs are more likely to be female, young, low-skilled, and single. People associated with a weaker labor market position are more likely to work informally than others. This is consistent with the literature (Gunther and Launov, 2012; Albertini and Terriau, 2019) and the view that salaried informal jobs can be a subsistence strategy for workers who lack opportunities in the formal sector. Finally, informal employees are also more likely to work less (41 h a week in formal work vs. 36 h in informal work) for a lower hourly wage (nearly 30% lower) than their counterparts in formal employment.

Figure 2 depicts the evolution of the Argentine labor market stocks between 2003Q3 and 2020Q1, as a percentage of the working-age population. First, let us notice that non-participation remains rather stable. This might suggest that nonparticipation plays a little role in accounting for unemployment changes. However, a given change in nonparticipation can actually be driven by large fluctuations in workers’ transitions in opposite directions, thereby canceling out each other. This problem, known as the “stock-flow fallacy” (Elsby et al., 2015), arises when using a stock-based analysis of the labor market. Our stock-flow framework will actually disentangle the role of fluctuations in transitions from/to nonparticipation in driving unemployment dynamics.

Figure 2

Second, employment in formal and informal jobs both exhibit an interesting pattern at the beginning of our sample. In 2003, formal and informal employment was on the rise until 2005Q1. From 2005Q2 onward, informal employment has been going down while formal employment kept expanding. Between 2012 and 2018, the Argentine labor stocks remained fairly stable. Since 2018, unemployment has started picking up, while formal employment and participation have started declining. In 2020Q1, the Argentine labor market shows early signs of deterioration.

Table 2 reports average transition probabilities between formal and informal employment, unemployment, and nonparticipation. The first striking feature concerns the labor market turnover. Unemployed workers face a total of 30.43% probability each quarter of finding a job, whether formal or informal, (which corresponds to an average duration on unemployment spell of 3.29 quarters). The average job separation rate in the informal sector amounts to 6.42% each quarter and 1.31% in the formal sector. For the sake of comparison, in the US, the total monthly job finding rate (40%) is more than twice as larger and the average unemployment duration (2.5 months) is about 7 months shorter (Shimer, 2012). In the US, the monthly job separation rate is around 3%, which is significantly larger than in Argentina. The lower turnover on the Argentine labor market calls for a methodology that accounts for changes in out-of-the-steady-state stocks to measure accurately worker flows and the variance decomposition of unemployment fluctuations (see Section 4.1).

A second striking feature is that informal work appears as an unstable labor market status. Inflows and outflows are several times larger than that of formal work. Each quarter, 6.42% of informal workers lose their jobs and 14% of them directly join the pool of nonparticipants. These outflows from informal work to nonemployment (whether to unemployment or nonparticipation) are five times larger than outflows from formal work (1.31% and 2.48% respectively). Finally, the chances of getting a job are higher in informal employment than in formal employment. Indeed, transition rates from unemployment to employment as a whole are largely explained by the transition rates from unemployment to informal employment. Specifically, the transition rate from unemployment to informal employment (23.69%) is much higher than that from unemployment to formal employment (6.74%), and from nonparticipation to informal work (7.74% per quarter vs. 1.82% to formal work).

Is informal employment a stepping stone to formal employment? Each quarter, 10.16% of informal workers become formal workers in the following quarter. While this rate involves a slow formalization process, it is still substantial when compared with the formal job finding probability from unemployment (UF at 6.74%) and nonparticipation (NF at 1.82%). Once the worker succeeds in getting a formal job, this labor market status appears very stable with a low quarterly separation rate of 1.31% to unemployment and 2.48% to nonparticipation. To that extent, we consider that informality introduces duality in the Argentine labor market, with fragile/unstable informal work and protected/stable formal work. This duality is reminiscent of the Spanish and French duality based on permanent versus temporary labor contract (Silva and Vazquez-Grenno, 2013; Cahuc et al., 2016).

Figure 3 summarizes the quarterly average worker flows between the four labor market states. With respect to Table 2, Figure 3 indicates the volume of quarterly flows in thousands of individuals. Table 2 is not informative enough on the magnitude of the worker flows as transition probabilities in the table apply to stocks of different sizes. In particular, as can be seen in Figure 2, the largest stocks are nonparticipation and formal employment. Over the whole period, the Argentine working population amounts to 25.8 million on average, with 9 million in formal employment, 5.2 million in informal employment, 1.6 unemployed workers, and 10 million out-of-the labor force.

Figure 3

Flows involving informal work are large: each quarter, 335,000 informal workers become unemployed, 746,000 of them leave the labor force. Flows in the reverse direction are also sizable: each quarter, 377,000 unemployed workers and 774,000 nonparticipants find an informal job. All inflows into informal employment involve 1,575,000 workers each quarter, while outflows from informal employment amount to 1,672,000 workers. Inflows are as large as outflows, which illustrate the “stock-flow fallacy”: a smooth evolution of stocks in Figure 2 actually hides large ins and outs. This illustrates why we need to look at worker flows to improve our understanding of the Argentine labor market.

We now decompose cyclical fluctuations in the unemployment rate into contributions attributable to each of the flow hazards. To do so, we adapt the dynamic decomposition of Elsby et al. (2015) to our empirical model. We obtain the following b statistic measuring the share of unemployment variance that is accounted for by the hazard rate from i to j:

(2)βij=CovΔut−1,t,Δu~t−1,tijVarΔut−1,t$$\beta^{i j}=\frac{\operatorname{Cov}\left(\Delta u_{t-1, t}, \Delta \tilde{u}_{t-1, t}^{i j}\right)}{\operatorname{Var}\left(\Delta u_{t-1, t}\right)}$$

where Δ is the first difference operator and u~t−1,tij$\tilde{u}_{t-1, t}^{i j}$the counterfactual unemployment rate obtained when only one worker flow fluctuates. To compute u~t−1,t,ij$\tilde{u}_{t-1, t,}^{i j}$we proceed as follows. First, we compute labor market stock changes that are driven by contemporaneous but also past changes in transition rates. This recursive formulation of stock variations is at the heart of the non-steady-state decomposition. Second, we express the variance of any given labor market stock as the sum of its covariance with any counterfactual obtained in the previous step. Note that in Eq. (2) a covariance appears in the expression of b such that negative values of b can possibly appear.

As we are not interested in the decomposition of stock changes per se but rather the decomposition of the unemployment rate, we use a first-order Taylor expansion to approximate unemployment changes:

(3)Δut=1−ut−1ΔUt−ut−1ΔEtLt$$\Delta u_{t}=\frac{\left(1-u_{t-1}\right) \Delta U_{t}-u_{t-1} \Delta E_{t}}{L_{t}}$$

with E_t being the total employment stock and L_t the labor force (the sum of U_t and E_t). Note that we should have ∑βij≈1$\sum \beta^{i j} \approx 1$where the difference from unity is accounted for by approximation errors. Similarly, changes in the informality rate are such that:

(4)Δit=1−it−1ΔIt−it−1ΔFtEt$$\Delta i_{t}=\frac{\left(1-i_{t-1}\right) \Delta I_{t}-i_{t-1} \Delta F_{t}}{E_{t}}$$

with i the informality rate and E_t the total employment stock (the sum of I_t informal employment and F_t formal employment).

Column (1) of Table 3 reports the shares of the variance of the unemployment rate accounted for by each hazard rate λ_ij. The originality of our contribution lies in documenting transitions to/from formal and informal work. Column (1) of Table 3 leads to stylized fact 1.

Fact 1, regarding fluctuations in the unemployment rate: nearly 40% of the fluctuations in the unemployment rate involve unemployment ins and outs from/to informal jobs (UI and IU). In addition, volatility in the exit rate from informal employment (IU) alone accounts for one-fourth of the unemployment rate’s variance.

Column (2) of Table 3 displays the variance decomposition of the share of informal work in total employment. This leads to fact 2.

Fact 2, regarding fluctuations in the informality rate: More than 40% of the variance of informality is driven by the variance of the formalization rate (transitions from informal to formal employment IF). Transitions between formal and informal employment (IF and FI) account for more than 60% of the variance of the informality rate.

Workers’ transition IF captures the stepping-stone effect or the formalization rate, that is, the ability of informal workers to get a formal job. Changes in this transition rate play a major role in the fluctuations of the informality rate in the Argentine economy.

The results also confirm the relevance of looking at labor participation when studying the Argentine labor market dynamics.

Fact 3, nonparticipation matters for the understanding of unemployment volatility but also for that of informality. 45% of unemployment volatility can be traced to the variance between nonparticipation from unemployment (UN and NU). Nearly 15% of volatility in the informality rate is driven by NI inflows to informal jobs from nonparticipants.

Finally, since the aggregate evidence points out the importance of nonparticipation, we further investigate this point by looking at labor market dynamics by gender as nonparticipation is known to be different across genders (See Figure 5, panel D). Male and female worker transitions probabilities are displayed in figure 6. Table 4 decomposes the contribution of the different flows for fluctuations in the male and female unemployment rates (columns (1) and (2)). Stylized fact 4 summarizes the main differences across gender.

Figure 4

Figure 5

Figure 6

Fact 4, regarding gender differences: transition involving nonparticipation matters more in the variance of female unemployment and informality rates than for their male counterparts. The main difference across genders is

• for the unemployment rate, the transitions from/to nonparticipation to/from unemployment NU and UN account for more than 50% of the female unemployment rate (vs. 35% for male unemployment rate)

• for the informality rate, nearly 20% of the variance in the informality rate is explained by changes in the inflows into informal jobs from female nonparticipants NI, versus 0% for male nonparticipants.

Fact 4 shows that Fact 3 is particularly relevant for female workers.

Interestingly, Facts 3 and 4 corroborate the idea of the added worker effect. It originally refers to an increase in the labor supply of married women when their husbands become unemployed (Mincer, 1962; Lundberg, 1985). As shown by Panel D of Figure 5, nonparticipation seems to evolve in opposite direction across genders: after the 2014 recession, more women left the labor market, their nonparticipation increased, while more men entered the labor market. Table 4 shows that female flows from nonparticipation account for a larger share of the variance in informality (NI) and unemployment (NU) than for men. Consequently, our results suggest the presence of an added worker effect that takes the form of larger transitions from nonparticipation to informal employment. In the case of Argentina, this effect seems to be confirmed by Martinoty (2014) during the 2001 Argentine crisis. Moreover, most studies point out the fact that it is deepened when households face credit constraints (Lundberg, 1985; Bingley and Walker, 2001) as it is the case in Argentina.

Fact 5 puts forward common features across genders: flows between unemployment and informality UI and IU explain nearly the same share (35-40%) of the unemployment rate volatility for men and women. The same occurs for the informality rate where more than 40% of the volatility of informal employment is explained by the sole exit rate from informality to formality IF.

Fact 5 shows that Facts 1 and 2 are relevant for male and female workers. Fact 5 also suggests that the flows from and into informality are important to understand unemployment fluctuations for men and women. For the informality rate, the formalization rate plays an important role in accounting for cyclical changes in informality, regardless of gender.