A version of this post was published in Business Day on April 12, 2023.

Unemployment, as we all know, is very high in South Africa and has been for decades. Mosomi and Wittenberg (2020) provide a good overview of the labour market over the last 20 years, and although unemployment did fall somewhat in the early years of the century - to a low of 21.5% in 2008 - it has steadily climbed since then. With the severe pandemic-induced recession of the last few years, South African unemployment is now slightly worse than its previous peak of 31.5% (in 2003). Most of this unemployment is long-term: Maluleke (2023) reports that just over 78% of the 7.8m unemployed people (narrowly defined) have been out of work for a year or longer.

It’s easy to imagine that at the individual level, there is nothing but unemployment, now and forever. In fact, this is not true: as in other countries, our economy is continuously reshaping itself, reallocating workers and jobs across firms and across sectors. One of the only papers to look at labour market dynamics in South Africa is Kerr (2018), using administrative data from all tax-paying employers, shows that “gross worker turnover” - the total number of hires and separations - amounts to about 50% of the workforce annually.

If this number seems high, remember that job-to-job transitions are, in a sense, counted twice. For example, suppose we have two firms with five workers each. If just one worker switches jobs with one at the other firm, there are two separations and two hires, and gross worker turnover would be 40%. Of course, in reality there are also many transitions are from employment to unemployment, or vice versa, that count towards this number. Kerr (2018) also points out that this level of “worker flows” is not out of line with those experienced in many OECD economies.¹

Even many workers who are by any reasonable definition disadvantaged will have some connection to the labour market. Dawson (2022) is an ethnographic account of the experiences of young black men, mostly without tertiary education, living in the townships of Johannesburg. Her interest is in how and why these young men often refuse low-wage jobs, or quit them once hired. Many of the men she interviews have had jobs in the formal sector, but for various reasons that she describes, they do not often stay in them for very long.

But this pattern of frequent transitions in and out of “marginal” jobs is actually quite common for developing countries, according to Donovan, Lu, and Schoellman (2023). In that paper, the authors assemble data from 49 countries on workers’ transitions in and out of employment. Among other things, they show that workers in developing countries have higher rates of entry and exit from employment than workers in rich countries. How is this consistent with the fact that unemployment rates tend to be higher in poor countries? The authors offer the metaphor of a “slippery ladder”: many workers find low-wage jobs easily, but do not progress to better-paying ones and often fall back into unemployment, or the informal sector.

Because it is relatively inexpensive to design interventions that affect the information available to workers and firms, there have been several high-quality experimental studies on job search and hiring in South Africa. Such interventions also create fewer logistical and ethical problems than, say, trying to retrain workers. For example, Abel, Burger, and Piraino (2020) show that providing job applicants with reference letters from previous employers increases the number of successful applications and eventual employment.² In a related paper, Abel et al. (2019) shows that helping job-seekers to plan their search does indeed raise the effectiveness of their search efforts. And Carranza et al. (2022) show that providing credible information about applicants’ skills (e.g. numeracy or communication) is also useful: it increases the number of applications sent by job-seekers, and subsequently their employment and earnings.

These are all exceptionally high-quality studies. Still, I don’t think these interventions, even if faithfully scaled up, can bring down unemployment by much. Below, I do a little arithmetic to explain why I’m skeptical.

Possibly more effective might be a version of the intervention of Bertrand and Crépon (2021). In that paper, the authors provide legal consulting services and training to a random subset of small and medium-sized businesses (the average number of employees, pre-intervention, was about 80). It seems many of these businesses had a poor understanding of labour law, and the treatment was effective in changing their perceptions of how costly or difficult hiring or dismissing workers would be. When the authors followed up with these firms six months after the experiment, the “treated” firms had expanded by, on average, 12% relative to the “control” firms. Unfortunately they do not study the effects on profitability - we’d hope that the induced growth was profitable, and therefore likely to be sustained over the long term. Still, simplifying labour regulations, or even just communicating them more effectively, seems like a promising direction for research or policy change.³

So what does explain the high and persistent level of unemployment? In my opinion, the obvious answer is only plausible one: supply exceeds demand at prevailing wages. The fact that kilometres-long queues of applicants show up for entry-level jobs is clear evidence of this.

In some cases, we also have independent evidence that manufacturing wages are high relative to a relevant comparison group - other countries. Gelb et al. (2020) compute “unit labour costs” - a measure of wages relative to labour productivity - for several countries; they find that almost all African countries in their sample look unfavourable by this metric, though South Africa ranks especially low (with Ethopia fairly high). One can question the construction of the productivity estimates in the denominator, but Li et al. (2012) show how these measures track the recent rise of labour costs in China. Since some types of low-wage manufacturing (e.g. clothing and textiles) are shifting out of China and to other Asian countries, these measures are likely picking up something real about labour demand, at least for tradeable manufactured goods.

The experiences of our peer economies should temper our expectations about what can be achieved with either informational interventions on the job search side, or even a more radical approach to deregulating labour markets and allowing formal-sector wages to fall. South Africa will not become Sweden anytime soon, and even if we succeed in bringing down unemployment by half, there will still be many “bad” jobs from which people will often quit.

Whether by accident or by design, our policymakers have chosen a combination of high wages and high unemployment, rather than lower levels of both. Perhaps this is the right choice, but I would like to hear someone defend it on those terms. Instead, the usual response is to deny that wages matter for labour demand, or to attack the integrity of anyone who suggests that they do. But markets cannot be browbeaten into submission. Like the wind and the tides, we can only adapt ourselves to their movements.

Appendix: Steady-State Calibration of Job Finding and Separation Rates

Let’s ignore the participation margin, and say the job-finding rate is \(f\) while the separation rate is \(s\). There is a unit mass of workers. The unemployment rate \(u_t\) obeys the law of motion

\[ \dot{u}_t = s\cdot (1-u_t) - f\cdot u_t. \]

Thus, the steady-state unemployment rate is

\[ \overline{u} = \frac{s}{s+f}. \]

Let’s take \(\overline{u} = 0.25\), about where unemployment has been, give or take a few percent, for the last 20 years. But to identify \(s\) and \(f\) separately we need another piece of information, which we could take from Kerr (2018)’s estimates of gross labour turnover. In this (entirely mechanical) economy, there are no job-to-job transitions, so total hires plus separations are \(f\cdot u_t + s\cdot(1-u_t)\) when the unemployment rate is \(u_t\). Thus we can take Kerr (2018)’s estimate that gross labour turnover is 50% of total employment to mean (in steady state)

\[ 0.5 = \frac{f\cdot \overline{u} + s\cdot(1-\overline{u})}{1-\overline{u}} = s + \left(\frac{\overline{u}}{1-\overline{u}}\right)\cdot f. \] Solving leads to \(s = 0.25\) and \(f = 0.75\). Now, how much could these job search interventions raise \(f\)? Carranza et al. (2022) estimate that their interventions raise employment probabilities by 5 percentage points. So if \(f\) rises to \(0.75 + 0.05 = 0.8\), the steady-state unemployment rate should fall to \(\overline{u} = 0.25/1.05 \approx 0.238\), i.e. a 1.2 percentage point reduction in unemployment. Better than nothing, I guess!

If you prefer the broad unemployment estimate of \(\overline{u} = 0.4\), repeating this exercise gives you “calibrated” values of \(s = 0.25\) and \(f = 0.375\). So a 5 percentage point increase in the job-finding rate leads to a new unemployment rate of \(\overline{u} = 0.25/(0.25 + 0.425) = 0.37\). Better - a 3 percentage point reduction in unemployment would be very welcome! - though off a higher base. Either way, I don’t think these type of interventions are going to really move the aggregate unemployment rate much.