KOBAYASHI Keiichiro: The problem of wage inequality has become a big topic in Japan only in the past two or three years. In the United States, however, wage inequality began to grow as early as the 1980s (possibly even the mid-1970s), and continues to grow to this day. Among economists, this long-running trend is seen as a phenomenon resulting from the skill-biased nature of technical change in the past several decades. As typified in the proliferation of computers, systematized techniques that require specialized skills (e.g., knowledge learned in university or other higher education) have more frequently become a condition for employment. Thus skilled workers have greater opportunities to increase their income while unskilled workers have lost jobs, having them replaced by computers.
To consider how the inequality problem will develop in the future, it is important to examine how skill-biased technical change has occurred and whether this change will continue. The paper I would like to introduce this time, written by Daron Acemoglu, provides many clues in thinking about these questions.
Daron Acemoglu (2002), "Directed Technical Change," Review of Economic Studies 69(4): 781-809.
Econo-kun: What is the paper about?
KOBAYASHI Keiichiro: The main thrust of Acemoglu's argument is that an increase in the supply of a particular production factor (skilled labor) due to exogenous forces creates the incentive (for companies) to develop technologies that favor skilled labor, leading to skill-biased technological progress. Consider capital, unskilled labor, and skilled labor as production factors and we can see that the spread of university education in the U.S. in the 20th century increased the amount of skilled labor (university-educated workers) relative to unskilled labor. Consequently, skill-biased technologies are developed and wages for skilled workers increase; an increase in the skill premium).
This theory adequately explains the rise in the university wage premium prior to the 1990s but its model cannot explain the fall of the middle class from the 1990s onward (in this period the number of jobs paying middle-class wages has reportedly decreased while the number of low-wage and high-wage jobs has increased; see Autor, Katz, and Kearney [2006]). Such shortcomings need to be kept in mind, but the ideas underlying "Directed Technical Change" offer important implications for the future directions of technical change.
Structure of Acemoglu paper
Though it takes the form of a neoclassical economic growth in format, Acemoglu's model can basically be considered a static model (the paper focuses solely on balanced growth paths and no consideration is given to transitional equilibrium paths). Skilled labor is described as Z, and unskilled labor as L. Consumer goods (which are also investment goods) are produced from two intermediate goods, YL and YZ, by using L and Z respectively, with the production process described by a CES (constant elasticity of substitution) production function.
The production of intermediate goods YL (or YZ) involves the use of L (or Z) and machines x that are complementary to L (or Z). Producers of intermediate goods are competitive companies. Thus, in determining the volume of production and the purchasing quantities of production factors, they take the prices of production factors and goods as given based on the recognition that such prices are determined by the market. There are NL varieties of machines for producing YL (NZ varieties for producing YZ), with NL and NZ representing the levels of labor-augmenting and skill-augmenting technologies respectively.
When simplified, the production function for intermediate goods can be described by a Cobb-Douglas production function.
Log YL = log NL + (1-β) log xL + βlog LLog YZ = log NZ + (1-β) log xZ + βlog Z
Machines are assumed to be invented by "technology monopolists" on a variety-by-variety basis and supplied to intermediate goods producers. (One single technology monopolist can lend multiple varieties of machines but cannot monopolize all varieties. Free entry exists for research and development [R&D] and machine rental activities and technology monopolists engage in monopolistic competition.)
Technology monopolists are presumably able to increase the variety of machines (invent new machines) steadily at a constant rate by investing consumer goods in R&D activities. Thus, the variety of machines, NL and NZ, changes with time; changes which represent technical change. A decrease in NL/NZ is termed as "skill-biased technical change."
The value of NL/NZ changes as a result of technology monopolists' optimization. In the case of a balanced growth path, NL/NZ is constant. Acemoglu statically examined how the value of NL/NZ relates to the ratio between Z and L (he assumes that Z and L are exogenously given).
Whether an increase in Z relative to L corresponds to skill-biased technical changes is determined by two factors: a price effect and a market size effect.
Price effect: An increase in Z induces an increase in YZ relative to YL, so that the price of YZ becomes lower than the price of YL. This reduces technology monopolists' incentive to produce Z-augmenting machines. The price effect serves to decrease NZ when Z increases.
Market size effect: An increase in Z enables an increase in the production volume of YZ relative to that of YL. This increases profit for the producer of YZ, bringing greater profit to the technology monopolist that invented the machine for producing Z. The market size effect serves to increase NZ when Z increases.
Model calculation results show that where Z and L are substitutable, the market size effect is greater than the price effect, thus an increase in Z results in a decrease in NL/NZ. Conversely, where Z and L are complementary, an increase in Z results in an increase in NL/NZ. (In cases where Z and L are strongly substitutable, when the elasticity of substitution is above 2, an increase in Z leads to a decrease in the relative wage WL/WZ, meaning the skill premium increases.)
While NL and NZ increase through R&D activities, the degree of increase may be affected by the current quantity of NL and NZ, in which case the innovation possibilities frontier is referred to as state-dependent. The greater the degree of state-dependence, the stronger the skill bias of technical change.
Acemoglu presents some examples of an actual economic phenomenon that can be explained by this model. Recently, skill-biased technical change has been occurring in the U.S. economy. According to the model, this is attributable to the spread of higher education in the U.S. in the 20th century and the resulting increase in the size of the university-educated labor force. The rise in the skill premium and increase in wage inequality between highly-educated and less-educated workers can also be explained by this model, provided that Z (skilled labor) and L (unskilled labor) are substitutable. The model also explains the reasons why technical change in the 18th and 19th centuries was "skill-replacing" (why the factory system favoring unskilled over skilled labor developed). In the 18th and 19th centuries, a large number of unskilled workers migrated from villages to urban areas, triggering a sharp increase in L. As a result, technical change grew biased in a way to further utilize L; skill-replacing technical change occurred.
By applying this model to analyze globalization (markets opening between developed and less developed countries [LDCs]), the following results can be obtained: trade liberalization induces technical change that is excessively skill-biased (for LDCs) and wage inequality between developed countries and LDCs expands (based on the premise that Z and L are substitutable).
This model can also explain the Habakkuk hypothesis that "the more rapid technical change or technology adoption in the U.S. economy during the 19th century relative to the British economy resulted from the relative labor scarcity in the U.S." Consider a production function that takes H (labor) and L (land) - instead of Z (skilled labor) and L (unskilled labor) - as inputs, and we can see that labor-augmenting technical progress should occur in the U.S., which is relatively labor-scarce and land-abundant, provided that H and L are substitutable.
Furthermore, the Acemoglu model properly explains the "wage-push shock" (wage increase) that occurred (or is perceived to have occurred) in continental Europe in the 1960s. It demonstrates that a wage-push shock will initially increase labor's share in national income but this will be followed by capital-augmenting technical change which increases the capital share and hence lowers the labor share. It also shows that employment will continue to fall throughout this process. This is roughly consistent with Europe's experiences.
Econo-kun: Using the Acemoglu model, can we explain what we see with growing inequality in Japan and polarization of the workforce in the U.S.? (Polarization refers to the labor market being split into some high-paying occupations and a large number of low-paying jobs with a decrease in middle-paying occupations for ordinary university and high school graduates. This trend reportedly became evident in the 1990s.)
KOBAYASHI Keiichiro: As discussed, a phenomenon such as polarization of the workforce cannot be explained by this model if we define skilled labor as university-educated workers and unskilled labor as those with a high school education or lower (because if so, wages for university-educated workers should have increased together with the progress of technical change).
But what if "skilled" is defined based on the relative importance of each type of skill in comparison with the dominant technology of the time. For instance, according to Acemoglu's explanation, farmers in the 18th century are regarded as unskilled labor. However, they should have been seen as skilled labor if defined in terms of agricultural technique. That is, because of the shift in the economy's dominant technology from agricultural to manufacturing technique, skilled agricultural workers became unskilled workers in the manufacturing sector.
Likewise, we can see the modern world as a society where the proliferation of computers is leading to manufacturing technology being replaced by information technology as the dominant technology in the economy. The "skilled labor" in the manufacturing-based society (e.g., university-educated workers) will turn into unskilled labor in the area of information technology. Accordingly, a substantial number of Z (skilled labor) are then reclassified as L (unskilled labor). This increase in L has perhaps been embodied by the polarization of the workforce in the 1990s or the growing inequality currently observed in Japan.
If this situation is so, L is increasing relative to Z. Thus the Acemoglu model would suggest that (skill-replacing) technical changes biased toward L (unskilled labor in the era of information technology) will proceed in the coming years. This may result in gradual narrowing of wage inequality over several decades, as seen in the U.S. and other developed countries from the late-19th through the mid-20th centuries. When we consider the future development of the problem of wage inequality from the standpoint of Acemoglu, wage inequality can be expected to not continue expanding in one direction but cycle through expansion and contraction over a very long term. In a period of transition characterized by a major change to the framework of technology, wage inequalities expand due to an increase in the population of workers unskilled in new systematized techniques. Yet at the same time this situation induces skill-replacing technical innovation. In due time, newly developed technology becomes accessible to these unskilled workers and wage inequality begins to narrow (unskilled workers turn into skilled workers). Later, when the number of skilled workers continues to increase, skill-biased technical changes occur and wage inequalities again begin to grow.
The Acemoglu model assumes that the supply of Z and L is exogenously given. However, when considering a long-term cycle such as the one mentioned above, it is necessary to assume a mechanism in which the supply of Z and L changes endogenously in response to technical change.
Lastly, there is another technical issue concerning the assumptions of the Acemoglu model. The model produces slightly different results depending on whether R&D is state-dependent (wherein the current level of technology [NL and NZ] has a spillover effect on the speed of technical innovation). According to Acemoglu, R&D is ideally state-dependent when explaining technical changes over the long term whereas state-dependence is not needed when explaining a short-term expansion in wage inequality. It might be possible to incorporate the question of state-dependence as an endogenous factor. For instance, if a patent is effective only for a finite period, we can probably develop a model in which R&D is not state-dependent in the short term but becomes state-dependent over the long term. Acemoglu acknowledges state-dependence as an area that needs to be further addressed. Thus, if we can develop a model in which R&D exhibits no state-dependence in the short run but is state-dependent in the long run, we can call this an improvement.
