A. Environment

The representative household maximizes utility under the given budget constraints as follows:

where C_t denotes consumption, I_t is investment, and Y_t is total output.

The law of motion for capital is expressed as

where K_t is capital stock and δ represents the rate of depreciation.

Final goods are produced by combining industry output using the CES aggregator, as follows:

An industry i produces industrial output using capital and labor, where labor is a composite of J occupations. Specifically, an industry i ’s output is given by

where A_i,t is industry i ’s productivity, K_i,t is industry i ’s capital stock, Z_i,t is industry i ’s labor composite, L_{i, j,t} is the labor supplied to industry i and occupation j , and M_j,t is occupation j ’s productivity. The parameter σ captures the elasticity of substitution across occupations (or tasks), and ν_ij is a weight parameter of occupation j used in industry i .

Note that ν_ij in equation (3) differs both by industry i and occupation j such that any change in occupation-specific productivity M_j has heterogeneous effects across industries as well. Similarly, a change in industry-specific productivity A_i also alters occupational employment because each industry employs labor at different levels of intensity.

B. Equilibrium

The final goods producer solves the following profit maximization problem:

where p_i is the price of industry i normalized by the price of the final goods; here I normalize the price of final goods to one.

Solving the final goods producer’s problem, we obtain

Each industry i ’s producer solves

where R is the rate of return on capital and 𝑤 is the effective wage rate per unit of labor.

From equation (2), a solution to the industry-level producer’s problem can be expressed as

where .

The market clearing conditions are

Finally, the representative household’s problem is expressed as

C. Model Solution

Given labor endowment L and capital endowment K , I compute the equilibrium allocation as shown below. For notational convenience, I denote industrial capital stock per capita as k_i (= K_i / L_i ) , industrial output per capita as y_i (= Y_i / L_i ) , and industrial labor composite per capita as z_i (= Z_i / L_i ) .

From equation (6), I have for all j . Therefore, we can express the occupational share in industry i ’s employment and industrial labor composite per capita as

From equations (5), (6), and (9), we have R / w = α_i /[(1 – α_i ) k_i] . Accordingly, the ratio of capital stock per capita between two industries ( i and I ) satisfies

Also, from equations (4) and (5),

Combining this with equation (2), we have the following expression for the ratio of industrial employment.

At this point, I can compute the ratio of industrial employment and industrial capital per capita from equations (11) and (12) and hence industrial employment and capital stock. Substituting industrial employment and capital stock into equation (2), we can compute industrial output ( Y_i ). Subsequently, we can compute the final output (Y ) from equation (1) and the industrial price ( p_i ) from equation (4). The equilibrium level of rate of return ( R ) and the wage rate ( w ) are obtained correspondingly from equations (5) and (6). Lastly, substituting industrial employment into equation (9), we find the occupational employment in each industry (L_ij ), from which occupational employment is determined, as follows:

Equations (9), (12), and (13) show how changes in exogenous productivity A_i and M_j affect industrial and/or occupational employment. For example, a rise in A_i would reduce the price of industry i , p_i . When the elasticity of substitution across industries is less than one (ε <1 ), the amount of input in industry i would become smaller. The first bracket in equation (12) shows this substitution effect.

Each industry employs all occupations with different levels of intensity, meaning that changes in occupational productivity (M_j ) also affect the industrial total factor productivity and hence industrial employment. For example, a rise in M_j would increase Ṽ_i more in an industry that employs occupation j more intensively than others (i.e., an industry with a higher ν_ij ). This would affect industrial employment through the second bracket on the right-hand side of equation (12). More formally, industry i ’s production is from equations (2) and (10), indicating that industry i ’s measured total factor productivity is , a combination of these values of A_i and M_j.

Similar to industrial employment, changes in both A_i and M_j affect occupational employment. Changes in M_j would alter occupational employment directly in equation (13) and indirectly through changes in industrial employment L_i . Because changes in A_i alter industrial employment, they also affect occupational employment, as shown in equation (13).

Note that an increase in would raise demand for occupation j (equation 13). Given that , an increase in is associated with an increase in M_j if σ >1 and a decrease in M_j if σ <1. Empirically, the literature finds elasticity of substitution across different occupations to be less than one, implying that an increase in occupation-specific productivity can be interpreted as technological progress substituting for labor in occupation j .

A. Estimation of production function parameters

The parameters of the final goods production function in equation (1) are the elasticity of substitution between industries (ε ) and the weight parameters ( γ_i ). From the equilibrium condition in equation (4), I formulate the following relationship:

I estimate the equation above by the iterated feasible generalized non-linear least squares (IFGNLS) method following Herrendorf, Rogers, and Valentinyi (2013). Because the substitution elasticity is greater than 0 and the weight parameters are located between 0 and 1, the estimation equation becomes

where the elasticity of substitution ( ε = 1 / (1 + e^b )) and the weight parameters ( γ_i = e^c_i / (1 + ∑e^c_i)) are inferred from estimates of b and c_i . The sample period is from 2005 to 2019, and the nominal and the real value added by economic activity from the National Accounts correspond to p_i Y_i and Y_i , respectively.

The estimation results are shown in Table 2. The estimated value of the elasticity of substitution between industries is 0.503 within the range of the values in previous studies that report complementarities ( ε <1 ) in one-digit industry classification schemes.

TABLE 2

ESTIMATION RESULTS: FINAL PRODUCTION

Source: Author’s calculations.

B. Calibration

The elasticity of substitution between occupations (σ ) governs how employment responds to a change in occupation-specific productivity (M_j ). Unfortunately, occupation-specific productivity (M_j ) and the elasticity of substitution between occupations (σ ) are not separately identified in our model. Therefore, I set the elasticity of substitution between occupations to 0.65, an average value of estimates in previous studies.1

Other parameters have been identified through the method of moments such that the data and endogenous variables of the model become similar in 2019. I calibrate the parameters to target the year 2019, not the average of 2010 to 2019, because one of the paper’s goals is to derive implications pertaining to structural changes over the medium run in the labor market after the pandemic. To do this, I assume that the year just before the pandemic represents the steady state and view the labor market after the pandemic as a transitional path from one steady state to another steady state. Note that this is somewhat different from the analysis of the business cycle, where average values over total sample periods are usually set as the steady state and where the analysis focuses on the short-run deviation from the steady state.

To be specific, I calibrate the values of ν_ij such that they match the employment share by occupation and by industry, the value of α_i to match the labor income share by industry, and the values of A_i,2019 to match the capital stock by industry as well as the level of aggregate output per total employment. Note that the model does not allow aggregate shocks to employment and the levels of total employment are given exogenously. That is, the model takes aggregate fluctuations as given and instead focuses on structural changes in the allocation of employment across industries and occupations. I therefore normalize the total number of workers in 2019 and the values of M_j,2019 to one. I then infer changes in the values of A_i , and M_j for the last decade (i.e., between 2010 and 2019) from the changes in employment by occupation and by industry. More specifically, I set the A_i,2010 to match industrial employment and the aggregate level of output in 2010 and the M_j,2010 to match occupational employment in 2010. I include detailed procedures for the calibration and data sources in Appendix A.

Table 3 summarizes the calibrated parameter values. The parameters for the occupational intensity levels (ν_ij ) reflect each industry’s employment structure by occupation. For example, the manufacturing industry features the highest fraction of routine workers compared to the services industries. Similarly, I can also confirm that the share of manual occupations is largest in the contact-intensive services industry, implying that a shock to routine occupations would disproportionately affect the manufacturing industry more, and a shock to manual occupations would have a more severe effect on contact-intensive industries.

TABLE 3

CALIBRATED PARAMETERS

Source: Author’s calculations.

Not surprisingly, manufacturing is the most capital-intensive sector (the highest α_i ). Among the service industries, the contact-intensive services sector is more labor-intensive than other services (α_contact < α_other ).

Between 2010 and 2019, the sector-specific productivity of the contact-intensive services sector declined most rapidly among the three broad sectors, but this should not be interpreted as a decline in total factor productivity, a combination of the sector-specific productivity and occupation-specific productivity rates. The manufacturing sector experienced slower growth in sector-specific productivity than other services, possibly indicating that the rate of the decline of manufacturing employment slowed after the Great Recession. In addition, I could confirm that routine occupations experienced the fastest growth in their occupation-specific productivity rates in an occupational dimension.

C. Model Fit

The employment structure in the model is set to be equal to the data in 2019 in terms of construction. On the other hand, the model and the data do not match precisely in 2010 because only industrial shocks (A_i ) and occupational shocks (M_j ) are allowed to change between 2010 and 2019. Therefore, by examining how similar the employment structures in the model and data are in 2010, I check how well the model explains the employment structure in Korea before the pandemic.

Figure 1 compares the employment share of the model with data by industry and occupation in 2010. The x-axis is the share of employment by industry and occupation in the data, and the y-axis is the share of employment by industry and occupation in the model. The dotted line is the 45-degree line. As shown in the figure, the employment shares in the model are very similar to the employment shares by industry and occupation observed in the data with an R-square value of.987, indicating that the model is suitable for an analysis of the employment structure in Korea. Again, aggregate variables should precisely match the data through the calibration procedure by construction.

FIGURE 1.

COMPARISON OF THE 2010 EMPLOYMENT SHARES BETWEEN THE MODELS AND THE DATA

Note: The x-axis is the share of employment by industry and occupation in the data, and the y-axis is the share of employment by industry and occupation in the model. The dotted line is the 45-degree line.

Source: Author’s calculations.

A. Identification of shocks

The employment shocks during the COVID-19 period have a form that shows changes in A_i and M_j . I identify A_i and M_j that match employment in 2020 and 2021 as follows:

1) Set the total number of employed persons in 2020 and 2021 to the data.

2) Set arbitrary values for M_{j ,t}.

3) Set arbitrary values for A_{I ,t} and k_{I ,t} .

4) Find k_{i ,t} based on k_{I ,t} and equation (11).

5) Find A_{i ,t} for 2020 and 2021 from the following equation:

where represents employment in industry i at time t in the data.

6) Iterate 3) to 5) over A_{I ,t} and k_{I ,t} until K_t is equal to the capital stock in 2019 and Y₂₀₂₀ / Y₂₀₁₉ (or Y₂₀₂₁ / Y₂₀₁₉ ) is equal to economic growth in the data.

7) Iterate 2) to 6) over M_j until , L_j,t / L_t in the model is equal to the data in 2020 and 2021.2

The procedure above produces A_i and M_j that match the thirteen industrial employment and eight occupational employment categories in 2020 and 2021 precisely. However, even if the thirteen employment by industry and eight employment by occupation categories coincide with the data, the detailed 104 (=13×8) employment cells by industry and occupation may not exactly coincide with the data. To check the accuracy, I compare the model with the data for the detailed 104 employment cells in Figure 2, finding that the model suitably explains the employment structure by industry and occupation during the pandemic. For example, the corresponding R-square outcomes between the occupational and industrial employment shares in the model and the data are 0.9975 and 0.9946 in 2020 and 2021, respectively.

FIGURE 2.

COMPARISON OF THE 2010 EMPLOYMENT SHARES BETWEEN THE MODELS AND THE DATA

Note: The x-axis is the share of employment by industry and occupation in the data, and the y-axis is the share of employment by industry and occupation in the model. The dotted line is the 45-degree line.

Source: Author’s calculations.

B. Characteristics of COVID-19 shocks

Although I identify the shocks that generate the employment dynamics during the pandemic, the economic meaning of these shocks is not straightforward. In the model, the only sources of exogenous variations are changes in A_i and M_j , which represent the productivity rates in each industry and each occupation. It would be natural to interpret these shocks as technological changes biased toward a certain industry or occupation when we focus on the long-run changes in the employment structure. However, there must be a much greater variety of shocks ongoing with regard to short-run fluctuations, such as markup, preference, and labor supply shocks, among others. Therefore, I would like to emphasize that the identified shocks herein should not be interpreted as structural sources of the variations in employment during the pandemic. Instead, the COVID-19 shocks identified herein should be understood as a combination of many structural shocks not explicitly reflected in the model.

However, the primary purpose of the identification of shocks is to gain an idea of which characteristics of an occupation or industry would be related to the observed changes in employment, rather than to delineate the contributions of various structural shocks on employment dynamics. For example, an occupation with higher infection risk would show lower employment caused by factors on both the demand and supply sides, and our exercise does not provide a clue as to exactly how much of the decrease in employment stems from a specific reason. Our exercise is still useful in that it defines the general nature of heterogeneity involved in the overall shocks to a certain occupation or industry, despite the fact that we do not know the contribution of each.

To provide economic implications with regard to COVID-19 shocks, I examine whether and how COVID-19 shocks are correlated with two variables that are suggested to be closely related to the pandemic in the literature: (1) the risk of infection and (2) easiness of remote work.

Recent studies utilize O*NET data to calculate the risk of infection index, and O*NET or the American Time Use Survey (ATUS) data to measure the ease of remote work (Adams-Prassl et al., 2020; Aum, et al., 2021c; Dingel and Neiman, 2020; Hicks et al., 2020; Mongey et al., 2021). O*NET asks experts and workers to give numerical answers to questions that capture detailed characteristics of an occupation, as defined by the Standard Occupation Classification (SOC) code. The ATUS data measure the amount of time people spend on various activities. In particular, it asks about “time worked from home,” which varies across industries as well as occupations.

I adopt the infection risk index and index for ease of remote work from Aum, Lee, and Shin (2021b) (henceforth work-from-home or wfh index) by industry and by occupation. The infection risk index is obtained using O*NET data examining the characteristics of each occupation in the US. Specifically, in O*NET, the degree of physical contact and exposure to diseases and infections are investigated and scored for each job. The infection risk index is the average value of two – the degree of physical contact and exposure to diseases and infections – after the standardization of each score. The work-from-home index is calculated using the weighted average of actual working at home in ATUS by industry and occupation. Finally, to match the indexes with our COVID-19 shocks, I assign US Census occupation codes and NAICS (North American Industry Classification System) to one-digit KSOC and KSIC.

Figure 3 shows the infection risk index against the work-from-home index by industry and occupation in Korea. Specifically, the x-axis is the work-from-home index (wfh) and the y-axis is the infection risk index (infect). The size of the circle represents the share of employment by industry and occupation in 2019. There is a negative (-) correlation between the two indexes, meaning that jobs with a lower risk of infection are generally more easily done at home. However, there is also considerable deviation from the regression line, implying that one index cannot completely represent the other and that the two indexes need to be examined separately. Aum, Lee, and Shin (2021b) also emphasized that the relationship between two indexes is far from tight, with an R-squared value only 0.034.

FIGURE 3.

INFECTION RISK AND WORK-FROM-HOME INDEX BY INDUSTRY AND OCCUPATION

Note: The x-axis is work-from-home index (wfh), and the y-axis is the infection risk index (infect). The size of the circle represents the share of employment by industry and occupation in 2019.

Source: Author’s calculations based on O*NET, ATUS, and EAPS.

I estimate the following regression to examine the relationship between COVID-19 shocks and the two indexes using employment by industry and occupation as a weight.

where wfh is the work-from-home index (by industry and by occupation), infect is the risk of infection index (by occupation), M_j,t is an occupation-specific shock, and A_i,t is an industry-specific shock. Note that β₁ > 0 if jobs with lower wfh outcomes (i.e., more difficult to do remote work) were hit harder by adverse employment shocks, and β₂ < 0 if jobs with higher infection risk were hit harder by adverse employment shocks. The regression analysis would guide us to a better understanding of the underlying sources of the variation in employment shocks by industry and occupation during the pandemic.

Table 4 shows the estimation results, which deliver three main results. First, in both 2020 and 2021, we have and , confirming the intuition that jobs that are more difficult to do remotely and with a higher risk of infection were hit harder both by occupation shocks and industry shocks, although these relationships were not always significant.

TABLE 4

RELATIONSHIP BETWEEN COVID-19 SHOCKS AND THE WFH OR INFECTION RISK INDEXES

Note: Standard error in parenthesis. *, **, and ** indicate significance at the 90%, 95%, and 99% percentiles, respectively.

Source: Author’s calculations.

Second, the relationships vary over time. In 2020, when employment rapidly declined, COVID-19 shocks show a significant correlation only with infection risk. However, the work-from-home index began to show a significant correlation with COVID-19 shocks in 2021, as employment began gradually to recover. On the other hand, the coefficient of infection risk ( β₂ ) becomes smaller in 2021 compared to this value in 2020.

Third, both indexes, risk of infection and work-from-home, have tighter relationships with COVID-19 shocks in an occupational dimension than in an industrial dimension. For example, the R-squared outcome in the regression with occupational shocks is 0.300, whereas that with industrial shocks is only 0.070. In addition, the t-value corresponding to the relationship between the work-from-home index and industrial shocks was 1.69, significantly smaller than with occupational shocks, at 7.28. This is not surprising given that the easiness of remote work is mainly related to the tasks a worker performs as opposed to the industry in which she/he works. I interpret this as meaning that occupational heterogeneity plays a more critical role in deriving the employment structure during the pandemic.

A. Future Technological Changes

The results in Section 4 suggest that COVID-19 raised the cost of employing contact-intensive tasks (specifically jobs for which the work-from-home is more difficult). According to the literature on directed technological change (e.g., Acemoglu and Restrepo, 2018, among others), an increase in the cost of employing contact-intensive tasks provides incentives to implement technological changes to replace these tasks. A natural question is whether such technological changes are feasible.

Although the literature has actively investigated the impact of technological changes on the labor market structure, it did not pay much attention to how contact-intensive each job is and whether new technology can replace contact-intensive tasks. However, even before the pandemic, recent technological changes enabled the replacement of contact-intensive tasks to some extent. For example, the widespread use of food-delivery applications through various platforms has replaced food-serving services in the restaurants with fewer workers. Many other examples indicate similar possibilities, such as the expansion of telemedicine due to the CPRSA Act in the US, the provision of online education services through the development of MOOC, or the development of smart-finance applications. Reflecting such trends, the Ministry of Science and ICT (2020; 2021) reports that the amount of O2O (online-to-offline) transactions in Korea grew by 22.3% in 2019, even before the pandemic, and its growth rate accelerated to 29.6% in 2020 through the pandemic. Therefore, I consider the acceleration of technological changes to replace contact-intensive tasks as a scenario that merits investigation.

B. Scenarios

The baseline scenario is that only past trends continue for five years, with no additional effects from COVID-19 appearing. I label this scenario as the baseline scenario, or Scenario 1.

Scenario 1. (past trends) For 2022 ≤ t ≤ 2026, occupation- and industry-specific productivity evolve as follows:

The expression above may seem complicated, but it merely says that the sector-specific and occupation-specific productivity rates grow at the average rate of growth between 2010 and 2019.

Compared to this scenario, I consider an alternative scenario in which the replacement of contact-intensive tasks will accelerate due to technological change biased toward contact-intensive task (CBTC) in the coming years. To reflect such technological changes, I assume that occupations with a lower work-from-home share will experience a more rapid increase in occupation-specific productivity compared to earlier trends. Hence, when new technologies replace contact-intensive tasks, occupations having more contact-intensive tasks will become relatively more productive. As shown in equation (13) in Section 2, a faster increase in an occupation’s productivity reduces its demand when occupations are complementary to each other (σ <1). Intuitively, firms allocate fewer resources to more productive tasks when tasks are complementary to each other.

To be specific, the alternative scenario (scenario 2) is expressed as follows:

Scenario 2. (past trends + COVID-19) For 2022 ≤ t ≤ 2026 , occupation- and industry-specific productivity evolve as follows:

Note that Scenario 2 is identical to Scenario 1 except for the m_j in the first equation. The additional term m_j captures CBTC, implying that the productivity rates of occupations that are difficult to do from home increase more rapidly. In other words, owing to CBTC (m_j ), M_j > M_l when 1 – wfh_j > 1 – wfh_l . Note again that the demand for occupation j falls when M_j becomes higher when σ <1 (equation 13).

The parameter τ governs the speed of CBTC and determines the distance between the productivity of the highest wfh_j and the productivity of the lowest wfh_j . I set τ = 0.14 , referring to the average speed of divergence between the highest M_j and the lowest M_j between 2010 and 2019, the pre-pandemic period.

Contrary to the previous section, the simulation exercise in this section is structural because I simulate a situation in which technological change is biased toward contact-intensive tasks. In other words, the simulation seeks to determine the structural effect of contact-intensive-task-biased technological changes on occupational and industrial employment rather than accurately to predict the employment dynamics after the pandemic. In this sense, I would like to clarify that the previous exercise does not provide evidence but suggests the possibility of CBTC. This is certainly a limitation of this analysis. Finding evidence of CBTC would require more data and analyses after the pandemic. This can be examined in future research.

C. Simulation Results

I simulate the model and the obtained equilibrium employment structures ( L_i,j,t / L_t ) under Scenario 1 and Scenario 2. Figure 5 depicts the occupational structures under the two scenarios. The line demarcated by the circle shows the observed employment share between 2005 and 2020. Not surprisingly, there was a declining trend in the routine employment share (-2.5%p between 2010 and 2021). Accompanying this trend, the cognitive employment share and manual employment share rose by +0.3%p and +2.3%p, respectively.

FIGURE 5.

OCCUPATIONAL STRUCTURE UNDER THE TWO SCENARIOS

Note: The line marked with circles is the observed employment share in the data; the black dotted line is Scenario 1, and the gray solid line is Scenario 2.

Source: Author’s calculations.

Similar to the continuous decline of routine employment before the pandemic, the routine employment share continues to fall by as much as -1.7%p for the next five years under the baseline scenario (solid gray line). At the same time, the cognitive employment share rises by +1.1%p, and the manual employment share rises by +0.6%p, a continuation of job polarization (or the declining trend of routine employment).

When the replacement of contact-intensive task accelerates (black dotted line, Scenario 2), however, the manual employment share changes, turning negative (+0.6%p → -0.3%p). A reduction in the demand for manual employment translates into greater demand for routine and cognitive employment than in previous trends, leading to a smaller decline of the routine employment share (from -1.7%p in Scenario 2 to -1.1%p in Scenario 2) and even higher increases in the cognitive employment share (from +1.1%p in Scenario 1 to +1.4%p in Scenario 2).

Equation (13) provides an idea as to why the replacement of contact-intensive tasks reduces the demand for manual workers. Because σ <1 in our calibrated model, an increase in M_j translates into a fall of , which leads to a reduction in L_j in Equation (13). Because m_j is higher for manual workers than for other occupations, manual workers experience lower demand compared to other types of employment. In other words, manual workers tend to have tasks that are difficult to complete from home (relatively lower wfh_j ) and hence are substituted more by technological changes that replace tasks that cannot be done at home.

I now turn to the industrial structure. Figure 6 shows the industrial employment structure under the two scenarios. In the data, the manufacturing employment share fell (-1.5%p from 2010 to 2021) and services employment increased through the process of structural transformation (circle-demarcated line). Within the service industry, the increase of line employment share focused on contact-intensive services (+2.6%p), whereas the share of other services (mostly high-skilled) experienced a decline (-1.1%p).

FIGURE 6.

INDUSTRIAL STRUCTURE UNDER THE TWO SCENARIOS

Note: The line marked with circles is the observed employment share in data; the black dotted line is Scenario 1, and the gray solid line is Scenario 2.

Source: Author’s calculations.

When the previous trend continues (Scenario 1), the manufacturing employment share decreases by -1.0%p over the next five years, while the employment share of contact-intensive services increases by +1.5%p and the employment share of other services falls by -0.5%p. However, as the replacement of contact-intensive tasks accelerates after the pandemic (Scenario 2), the demand for contact-intensive services will be reduced by -0.3%p over the next five years a -1.8%p reduction from +1.5%p in Scenario 1. As the demand for contact-intensive services decreases, the decline in manufacturing will ease, shifting from -1.0%p in Scenario 1 to -0.003%p in Scenario 2. Also, the demand for other services will boost the employment share of these workers by +0.3%p.

Discussion

The simulation results demonstrate that the acceleration of contact-intensive replacement technological changes (or CBTC) would alleviate the structural changes in employment observed in the past, such as job polarization. This would occur because jobs with more significant portions of contact tasks (i.e., remote work being more difficult) are different from jobs that involve routine tasks, which have been replaced heavily by earlier technological changes, also known as the IT revolution.

It is important to note that these results should not be interpreted as meaning that routine tasks will not be replaced in the future. Instead, they suggest that a broader range of jobs, both routine tasks and contact-related tasks, may be in danger after the pandemic given the more recent technological changes that have occurred. I highlight the potential acceleration of the former type of automation due to the pandemic and present related implications with regard to occupational or industrial employment structures.

The CBTC scenario (Scenario 2) includes technological changes reflecting past trends as well as the acceleration of contact-intensive task replacement. Although a contact-intensive task is different from a routine task, they are not mutually exclusive. In other words, jobs intensive in routine tasks may or may not be intensive in contact tasks. For a more precise interpretation, I classify jobs by both routineness and contact-intensiveness in Table 5. I classify jobs with a work-from-home index below average as contact-intensive jobs and those with a work-from-home index above average as non-contact-intensive jobs. This classification of routine and non-routine jobs follows the literature, e.g., Acemoglu and Autor (2010).

TABLE 5

OCCUPATIONAL CLASSIFICATION BY ROUTINENESS AND CONTACT-INTENSIVENESS

Note: 1) Contact-intensive jobs are those for which the wfh index is below average, and non-contact-intensive jobs are those for which the wfh index is above average, 2) The classification of routine and non-routine jobs follows Acemoglu and Autor (2010).

Before the pandemic, a widely accepted view with regard to the labor market structure was that routine jobs had disappeared, regardless of their degree of contact-intensiveness. Our simulation is based on the possibility that contact-intensiveness can serve as an additional dimension of future technological changes due to the COVID-19 pandemic. Therefore, among routine jobs, the share of routine and contact-intensive jobs will decrease further, adding to the previous decline. In addition, the share of non-routine and non-contact-intensive jobs will increase more rapidly than in the past. Most interesting is that the demand for non-routine and contact-intensive jobs, i.e., manual jobs, shifts from increasing to decreasing with the widest gap between the two scenarios, as highlighted in Section 5.C.

Table 6 compares the employment composition of manual, routine, and cognitive jobs in 2020 in Korea. Manual jobs have a higher proportion of temporary and daily workers than other jobs (45% vs. 15% or 10%), and the share of low-educated (up to high school) workers is also higher than in the other cases (78% vs. 56% or 13%). Meanwhile, more elderly people (over age 60) work manual jobs than in other cases (31% vs. 18% or 6%), meaning means that the reduced demand for manual workers due to the pandemic will burden mostly socio-economically vulnerable workers, which calls for policies supporting vulnerable groups in the labor market.

TABLE 6

LABOR FORCE COMPOSITION BY OCCUPATIONAL GROUPS IN 2020

Source: Economically Active Population Census (2020).