Productivity and Employment in a Developing Country: Some Evidence from Korea (original) (raw)

Productivity and Employment in a Developing Country: Some Evidence from Korea

SANGHO KIM
Honam University, Gwangju, South Korea
HYUNJOON LIM
Bank of Korea, Seoul, South Korea
and
DONGHYUN PARK*
Asian Development Bank, Mandaluyong City, Metro Manila, Philippines

Abstract

Summary. - The vast majority of the sizable empirical literature which examines the relationship between productivity and employment examines data from developed countries. In this paper, we contribute to the limited empirical literature on the productivity-employment relationship in developing countries by applying structural vector autoregression (VAR) models on Korean data. We find that produc-tivity-enhancing technology shocks reduce hours worked in the short run. Such evidence is qualitatively similar to findings from developed countries, and more consistent with sticky price models than the real business cycle theory. Although productivity-enhancing technology shocks are an important source of economic growth in Korea, they may have exerted a negative impact on employment. © 2009 Elsevier Ltd. All rights reserved.

Key words - technology shocks, productivity, employment, sticky prices, Keynesian, real business cycle, Korea

1. INTRODUCTION

Traditional Keynesian theory emphasizes the central role of demand-side factors such as monetary, fiscal, and investment shocks in macroeconomic fluctuations. In contrast, real business cycle (RBC) theory puts forth technology shocks as the main drivers of business cycles. A major prediction of RBC theory is a high positive correlation between productivity and employment. The underlying idea is that a positive technology shock increases both productivity and demand for labor, which, in turn, increases employment. Unfortunately for RBC theorists, a well-known stylized fact from US data-no correlation and indeed often negative correlation between productivity and employment-has led many economists to question the relevance of their theory. A substantial literature has recently emerged to empirically examine the relationship between productivity and employment more rigorously.

The pioneering paper by Gali (1999) finds that productivityenhancing technology shocks reduced hours worked in the US as well as all other G7 economies except Japan. The substantial body of research that confirms and supports Gali’s milestone findings include Basu, Fernald, and Kimball (2006), Francis and Ramey (2005), Francis, Owyang, and Theodorou (2003), Gali (2004), Gali and Rabanal (2004), Shea (1999) and Kiley (1998). A number of studies have challenged the robustness of such evidence, primarily on methodological grounds. These include Christiano, Eichenbaum, and Vigfusson (2003), Uhlig (2004), Dedola and Neri (2004), Peersman and Straub (2004), Chang and Hong (2006), and Chang, Hornstein, and Sarte (2006). In any case, a negative effect of productivityenhancing technology shocks on employment cannot be rec-
onciled with standard versions of RBC models and is more consistent with the sticky prices of Keynesian models. 1{ }^{1} The basic idea is that price rigidity prevents demand from changing in the face of lower marginal costs due to productivity gains, leading firms to produce same output with less labor.

The central objective of our paper is to empirically investigate the effect of technological shocks on productivity and employment in Korea. Therefore, one contribution of our paper is to re-examine the relationship between productivityenhancing technology shocks and employment using Korean data. However, our more significant contribution is to use data from a developing country to investigate this relationship. The vast majority of the existing empirical literature on the relationship between productivity and employment looks at data from the US and other developed countries. The limited number of studies on RBC in developing countries includes Mendoza and Smith (2006), Carmichael, Keita, and Samson (1999), and Chyi (1998). Studies on RBC in Korea are similarly limited and includes Yoon (2006), Park (2000), and Masih and Masih (1995).

However, neither set of studies looks at the technologyemployment relationship or seeks to otherwise test for RBC theory. That is, those studies look at issues other than how technology shocks affect productivity and employment or, more generally, how such shocks drive the business cycle in Korea and other developing countries. Yet the relationship between technology shocks on the one hand and productivity,

[1]

1. • We would like to thank the editor, Oliver T. Coomes, and four anonymous referees for their helpful comments and suggestions, although all errors are our responsibility. Final revision accepted: October 21, 2009.
     ↩︎

employment, and the business cycle on the other hand is no less important in developing countries than in developed countries. To the contrary, technological shocks may play a bigger role in developing countries, due to their relative technological backwardness and hence greater scope for technological progress. This is especially true for Korea, where imported foreign technology played a central role in the country’s rapid industrialization and economic growth. While the role of technology in long-term economic growth has long been recognized and studied, there has been very little research on the role of technological shocks on the business cycles of developing countries, as noted above. We hope that our paper will help to shed some light on the sources of business cycles in developing countries and thus contribute to the limited empirical literature on the topic. To do so, we empirically investigate the effect of productivity-enhancing technology shocks on Korean employment.

The rest of our paper is organized as follows: In Section 2, we describe the basic framework for our empirical analysis. More specifically, we describe the construction of our data, along with the bivariate structural vector autoregression (VAR) model of productivity and hours worked we use to identify technology shocks. In Section 3, we report and discuss the results of our estimation of the bivariate model. We also introduce another variable-prices-to build a trivariate structural VAR model, estimate it, and report and discuss the results. In Section 4, we draw conclusions and policy implications from our main empirical findings.

2. BASIC EMPIRICAL FRAMEWORK

In this section, we lay out our basic empirical framework. We explain why we choose total factor productivity (TFP) as our measure of productivity as well as how we construct our TFP data. We also describe how we plan to identify technology shocks by using bivariate structural vector autoregression (VAR) model.

(a) Data construction

Many empirical studies of the employment-productivity relationship use labor productivity as the measure of productivity, but this partial measure fails to take into account factor substitution between capital and labor. Such substitution is especially important for the Korean economy, which has continuously experienced capital deepening and adoption of new production technologies. Labor productivity generally depends on capital deepening as well as technological progress and structural efficiency changes. Furthermore, it is often argued that Korean economic growth has been driven mostly by factor accumulation rather than by productivity growth. In light of these facts, we use total factor productivity (TFP), which incorporates the effects of both structural and technological changes, as well as labor productivity as our
productivity measures. 2{ }^{2} Given Korea’s remarkably rapid economic growth and the central role of physical capital accumulation in that growth, it seems intuitively plausible that the country’s capital-labor ratio grew rapidly and has not been stationary. Empirical evidence provides some support to the non-stationary nature of the capital-labor ratio in Korea. 3{ }^{3}

The data for labor productivity, which is defined as the ratio of gross domestic product (GDP) to total hours worked, are compiled by the Bank of Korea. We constructed our TFP data from various sources in the Bank of Korea database and used the data to estimate Solow residuals for the period 1985Q12003Q4. The capital stock is the real amount of tangible fixed assets, adjusted for the capital utilization rate. Our measure of employment is hours worked, which we derived as the product of total number of employed workers and average hours worked per week (hi)\left(h_{i}\right). All variables other than hours worked are converted into constant 2000 prices.

The use of the Solow residual as a measure of productivity growth implicitly assumes perfect competition, constant-re-turns-to-scale technology, and full employment of labor and capital. More realistically, if any of these assumptions are violated, for example if there is less than full employment of capital and labor, the Solow residual may be affected by demandside variables and thus become an inaccurate measure of productivity growth. 4{ }^{4} In the case of Korea, Kim and Lim (2004) find empirical evidence that the Solow residual is not a strictly exogenous variable but instead co-moves with demand shocks. If measured productivities are influenced by the business cycle, a correlation between productivity and employment may be spurious and due to a correlation between employment and the business cycle. For this reason, it is desirable to control for cyclical bias in the productivity measure.

To address this problem, we follow the method suggested by Basu and Kimball (1997) and Ball and Moffitt (2001). To eliminate cyclical effects from the measured Solow residual, we regressed the log differences of the measured Solow residual on the log differences of the composite index (CI) for business cycles. We then adjusted the averages of the regression error terms to be equal to the original productivity measures, after controlling for cyclical effects. To address the endogeneity problem, we used the generalized method of moments (GMMs) with 1- and 2-period-lagged CI as well as M2 and 1-period-lagged M2 growth as the instruments. Figure A1 in the Appendix shows the growth rates of measured Solow residual and TFP estimates we obtained after eliminating the cyclical effects from the residual. TFP grew rapidly after the mid-1980s but slowed somewhat in the 1990s, and collapsed during the crisis of 1997-98. TFP recovered from the crisis shortly thereafter but then fell again after 2000.

Table 1 presents the lagged correlation coefficients between the variables used in our empirical analysis. Real GDP has coefficient values of 0.54 and 0.48 with hours worked and employment, respectively. TFP is highly correlated with hours worked and the GDP, leading the two variables by a quarter or showing simultaneous correlation. The correlation coefficients

Table 1. Lag correlation coefficients between key variables in Korea, 1985Q1-2003Q4

Notes: TFP denotes total factor productivity. Lag correlation coefficients measure correlation between the former variable in the present period and the latter variable in the t±it \pm i period.

support the predictions of RBC theory in the sense that productivity growth is closely related to macroeconomic fluctuations. Based on this observation, in the next section, we apply a structural VAR model to Korean data to investigate the dynamic relation between technology shocks and employment.

Prior to our empirical analysis, we carried out augmented Dickey-Fuller (ADF), Phillips-Perron (PP), and Kwiatkowski, Phillips, Schmidt, and Shin (KPSS) unit root tests to examine whether the time-series of the variables follow stochastic trends. Table A1 in the Appendix reports the test results for both levels and first differences. The tests unambiguously suggest the existence of one unit root for every variable, indicating that the time-series are integrated of order 1, I(1). We performed Johansen’s cointegration test on various sets of variables to check for the existence of a long-run rela-
tionship among the variables. The results, which we report in Table A2 in the Appendix, indicate that there is no cointegration vector and thus no long-run time-series relationship among the variables.

(b) Bivariate structural vector autoregression (VAR) model

In view of the absence of a cointegrating relationship among the variables, we specify a bivariate structural vector autoregression (VAR) model of TFP and hours worked to identify technology shocks in the Korean economy. While Shea (1999) used the number of patents and R&D expenditures as proxies for technology shocks, Gali (1999) used the long-run restriction that only technology shocks can affect productivity permanently in a structural VAR model. Although Shea’s

Figure 1. Cumulative impulse responses of labor productivity and hours worked to technology and non-technology shocks in Korea, 1985Q1-2003Q4.

Figure 2. Cumulative impulse responses of total factor productivity (TFP) and hours worked to technology and non-technology shocks in Korea, 1985Q12003Q4.

Figure 3. Cumulative impulse responses of adjusted total factor productivity (TFP) and hours worked to technology and non-technology shocks in Korea, 1985Q1-2003Q4.
method may be able to solve some measurement problems, such as those associated with procyclical movements of productivity, it cannot replace Gali’s identification method due to the low explanatory power of the proxies. 5{ }^{5}

Let ztz_{t} be a vector of TFP growth (xt)\left(x_{t}\right) and total hours worked growth (ht),zt=∣Δxt,Δht∣2\left(h_{t}\right), z_{t}=\left|\Delta x_{t}, \Delta h_{t}\right|^{2}, and εt\varepsilon_{t} be a vector of log of technology shocks (εts)\left(\varepsilon_{t}^{s}\right) and log of non-technology shocks (εth)\left(\varepsilon_{t}^{h}\right), εt=[εts,εth]2\varepsilon_{t}=\left[\varepsilon_{t}^{s}, \varepsilon_{t}^{h}\right]^{2}. Then, the kk-lag VAR of TFP growth and hours growth can be written as
Φ(L)zt=εt\Phi(L) z_{t}=\varepsilon_{t},
where Φ(L)\Phi(L) is a kk th-order matrix polynomial in the lag operator. The VAR can be rewritten in its moving average (MA) representation:
zt=C(L)εtz_{t}=C(L) \varepsilon_{t}
where C(L)C(L) is an infinite polynomial matrix in the lag operator Φ(L)=C(L)−1\Phi(L)=C(L)^{-1}. We can rewrite (2) as
zt=[C11(L)C12(L)C21(L)C22(L)][εtsεtsu]z_{t}=\left[\begin{array}{cc}C_{11}(L) & C_{12}(L) \\ C_{21}(L) & C_{22}(L)\end{array}\right]\left[\begin{array}{c}\varepsilon_{t}^{s} \\ \varepsilon_{t}^{s u}\end{array}\right].
Each of the elements is a polynomial in the lag operator. Two disturbances of technology and non-technology shocks cause fluctuations in TFP and hours worked, and are assumed to be orthogonal to each other. 6{ }^{6} To identify the technology shock, εts\varepsilon_{t}^{s}, we impose the long-run restriction that the non-technology shock’s long-run impact on productivity be equal to zero. This implies that C12(L)=0C_{12}(L)=0 and restricts the unit root in TFP to originate solely from technology shocks. C11(L)C_{11}(L), C21(L)C_{21}(L) and C22(L)C_{22}(L) refer to the long-run impact of technology shock on productivity, long-run impact of technology shock on hours worked and long-run impact of non-technology shock on hours worked, respectively.

3. EMPIRICAL RESULTS

We report and discuss the results of our estimation of the bivariate VAR model outlined in the preceding section. We
also introduce an additional variable-prices-to specify a trivariate structural VAR model, estimate it, and report and discuss the results.

(a) Results of the bivariate structural VAR model

In this sub-section, we report the results of our estimation of the bivariate model of productivity and hours worked described in (3) above. We chose the lag length of four to minimize the Akaike Information Criterion (AIC), Schwarz Criterion (SC) or Bayesian Information Criterion (BIC).

Table 2. Decomposition of forecast error variance of adjusted total factor productivity (TFP), hours worked and GDP deflator for Korea, 1985Q12003 Q 4

However, changing the lag length does not affect our results. 7{ }^{7} We first define productivity as labor productivity, as in Gali (1999). Figure 1 shows the cumulative impulse response of labor productivity and hours worked to technology and nontechnology shocks in the bivariate model. The responses are defined in terms of the natural logs of the levels rather than growth rates of the endogenous variables. The standard errors and confidence intervals are computed by bootstrapping 1,000 random draws. Labor productivity rose permanently to higher levels after initial adjustments in response to a one-standarddeviation positive technology shock.

The response of hours worked to technology shock was negative but insignificant. Our finding of negative but insignificant effect of productivity-enhancing technology shocks on hours worked is qualitatively very similar to Gali (1999). As noted earlier, such evidence casts doubt on the validity of RBC theory and is more consistent with price rigidity, which is a cen-
tral assumption of Keynesian models. Sticky prices prevent demand from adjusting in the face of lower marginal costs and thus encourage firms to produce the same amount of output with less labor.

Figure 2 shows the impulse response of the bivariate model after we replaced labor productivity with TFP as our measure of productivity. The confidence interval is computed by bootstrapping 1,000 random draws. The most striking feature of Figure 2 is that the response of hours worked to technology shocks is negative rather than positive on impact. Such evidence is more supportive of Keynesian-type sticky price model than RBC models. In fact, hours worked did not show a positive and significant response to a positive technology shock until the second quarter. The effect of positive non-technology shock on TFP was statistically insignificant, even in the short run. This result is consistent with the assumption that TFP is statistically orthogonal with non-technology shocks such as

Figure 4. Cumulative impulse responses of adjusted total factor productivity (TFP), hours worked and prices to technology and non-technology shocks in Korea, 1985 Q1-2003Q4.

demand or mark-up shocks, even in the short run. All other results, including the permanent increase in TFP to higher levels, are qualitatively similar to the results we obtained using labor productivity instead. 8{ }^{8}

As pointed out in the previous chapter, the estimated Solow residual may be an imperfect measure of total factor productivity in the presence of cyclical effects. To eliminate the cyclical effects, we adjusted the Solow residual by using a composite index of business cycles and demand-related instrumental variables. Figure 3 shows the impulse responses of the bivariate model when we used the adjusted TFP‾\overline{T F P} as our measure of productivity. They are generally similar to the responses we obtained earlier when we used the unadjusted TFP as our productivity measure. In particular, as was the case for Figures 1 and 2, the initial response of hours worked to technology shocks is negative rather than positive. Our finding that technology shocks have a negative impact on hours worked is more supportive of sticky price models than RBC models. Hours worked began to show a positive and significant response to positive technology shocks only in the second quarter (see Table 2). 9{ }^{9}

(b) Trivariate structural VAR model and its results

So far we have used the bivariate structural VAR model of productivity and hours worked to investigate the impact of technology shock on employment in Korea. Our estimation results indicate that positive technology shocks have a negative short run impact on employment, and are thus more consistent with sticky-price Keynesian models than RBC theory. The bivariate model lumped together all shocks other than technology shocks as non-technology shocks. These include demand shocks such as monetary policy, fiscal policy or shifts in business confidence, mark-up shocks associated with changes in oil prices, other input prices or terms of trade, and labor supply shocks. Since it is unlikely that any of these diverse shocks affect productivity in the long run, the long-run restriction we use in our model remains appropriate.

Nevertheless, decomposing non-technology shocks may be helpful for a more in-depth analysis. For example, dividing non-technology shocks into labor supply shocks and price shocks allows us to analyze their effects on employment, output, and prices. Price shocks generally reflect demand shocks. We now expand our bivariate model into the following trivariate model:
[ΔTfΔhtΔpt]=[C11(L)C12(L)C13(L)C21(L)C22(L)C23(L)C31(L)C32(L)C33(L)][cttctts−tctts−p]\left[\begin{array}{c}\Delta T_{f} \\ \Delta h_{t} \\ \Delta p_{t}\end{array}\right]=\left[\begin{array}{lll}C_{11}(L) & C_{12}(L) & C_{13}(L) \\ C_{21}(L) & C_{22}(L) & C_{23}(L) \\ C_{31}(L) & C_{32}(L) & C_{33}(L)\end{array}\right]\left[\begin{array}{c}c_{t}^{t} \\ c_{t}^{t s-t} \\ c_{t}^{t s-p}\end{array}\right],
where ΔTf,Δht\Delta T_{f}, \Delta h_{t}, and Δpt\Delta p_{t} denote adjusted TFP‾\overline{T F P} growth, hours worked growth, and GDP deflator growth, respectively, cttc_{t}^{t} denotes technology shock, and ctts−tc_{t}^{t s-t} and ctts−pc_{t}^{t s-p} denote the two nontechnology shocks-labor supply shock and price shockrespectively. Cij(L)C_{i j}(L) represents the long-run multipliers of the shocks on the endogenous variables.

Identifying the trivariate structural VAR model requires three restrictions, along with symmetry and normalization conditions of the covariance matrix of error terms, Ω=Var⁡(εt)\Omega=\operatorname{Var}\left(\varepsilon_{t}\right). We follow Blanchard and Quah (1989) in assuming that there are two types of disturbances-demand disturbances which have no long-run effect on output and supply disturbances which may have a long-run effect on output. Due to nominal rigidities, demand disturbances have effects on output in the short run but those effects fade in the long run. Only supply disturbances affect output in the long run.

The Blanchard and Quah assumptions imply two long-run restrictions for our purposes. First, they allow us to retain our earlier restriction that only technology shock can affect productivity in the long run. Therefore, the two non-technology shocks, namely labor supply shock and price shock, have no impact on long-run productivity-That is, C12(L)=C13(L)=0C_{12}(L)=C_{13}(L)=0. Second, demand or price shocks do not affect hours worked in the long run, implying the long-run restriction of C23(L)=0C_{23}(L)=0.

Figure 4 shows the impulse responses in the trivariate structural VAR model. As was the case for Figures 1, 2, and 3, hours worked slightly fell at first in response to a positive technology shock. Such negative short-run response is more supportive of Keynesian-type sticky price models than RBC models. However, hours worked then started to rise within a quarter. In response to a positive labor supply shock, hours worked rose at first, then fell before reaching its new equilibrium after ten quarters. In response to a positive demand or price shock, which was assumed to have no long-run effect on hours worked and productivity, hours worked rose slightly at first but returned to its initial level after two years. The GDP deflator increased rapidly for 12 quarters in response to a positive demand shock and kept increasing modestly thereafter. In contrast, the GDP deflator fell in response to positive technology and labor supply shocks. However, the response of the GDP deflator is insignificant for the first few quarters.

4. CONCLUDING REMARKS

According to real business cycle (RBC) theory, the business cycle is driven largely by technology shocks rather than the traditional Keynesian demand shocks associated with macroeconomic policy or business confidence. A major empirically testable prediction of RBC theory is a positive relationship between productivity and employment. Troublingly for RBC advocates, a substantial empirical literature initiated by Gali (1999) finds that productivity-enhancing technology shocks reduced employment in the US and other developed countries. Although a number of studies challenge the robustness of this literature, the balance of evidence seems more supportive of a negative relationship than a positive relationship. This has cast serious doubt on the empirical validity of RBC theory among many economists.

In this paper, we re-examined the relationship between pro-ductivity-enhancing technology shocks and employment using quarterly Korean data. More specifically, we used a bivariate structural VAR model of productivity and hours worked with two types of shocks-technology and non-technology-along with the long-run restriction that non-technology shocks cannot permanently affect productivity. Our empirical results show a negative but an insignificant effect of positive technology shocks on hours worked when we used labor productivity as the measure of productivity. Furthermore, even when we replaced labor productivity with total factor productivity (TFP) as our productivity measure, we found that technology shocks had a negative effect on hours worked on impact and in the very short run. This finding lends more support to Keyne-sian-type sticky price model than to the presence of a real business cycle. We were able to replicate this finding-that is, the absence of a positive effect on hours worked in the very short run-when we adjusted our measure of total factor productivity, the Solow residual, to control for cyclical effects.

On the other hand, we find a positive effect of technology shocks on hours worked in the medium and long runs, and it

is possible to interpret this as evidence of RBC models. However, in the medium and long horizons, the response of hours worked to technology shocks cannot meaningfully distinguish between the RBC and sticky price models. The underlying reason is that prices are more flexible and thus adjust beyond the short run. In particular, if prices are set one period in advance, prices may fall after the second period in response to a positive shock. As a result, hours worked may increase rather than decrease because of an increase in real balances. Therefore, a positive impact of technology shocks on hours worked in the medium and long run is consistent with sticky price model. In terms of distinguishing between the two models, what really matters is the response of hours worked on impact and in the very short run and, as noted above, in this respect our evidence is more favorable to sticky prices than RBC.

We then added another variable, the overall price level, to expand our bivariate model to a trivariate model of productivity, hours worked and GDP deflator. We divide non-technology shocks into labor supply shocks and demand or price shocks. Our empirical results re-confirm a negative effect of productivity-enhancing technology shocks on hours worked on impact. As was the case for our bivariate model, however, the effect turned positive and significant beyond the short run. The response of the GDP deflator to technology shocks was insignificant in the short run. All in all, in the case of Korea, our findings fail to provide convincing support for RBC models and are, if anything, more consistent with sticky price models in light of the negative response of hours worked in the very short run.

Our results for Korea are thus qualitatively similar to those from the results of earlier studies for developed countries, in particular Gali (1999). According to our evidence, although technological progress has been an important source of longrun growth in Korea, its impact on hours worked has been negative in the short run. Given that much of advanced foreign technology came into Korea in the form of imported capital goods, there may have been substitution between capital and labor over time. Such interpretation is consistent with the fact that the Korean economy has evolved from a laborintensive production structure to a capital- and technologyintensive production structure. The implication for other developing countries is that while technological progress promotes economic growth it is unlikely to contribute to employ-
ment growth in the short run. Put differently, our evidence indicates that the risk of jobless growth driven by technology shocks is relevant not only for developed countries but developing countries as well. Since job creation is a pressing socioeconomic concern in developing countries, our finding implies that policymakers should take measures to minimize potential short-run job losses arising from technology shocks.

A significant contribution of our study to the literature is to use data from a developing country to look at the relationship between productivity-enhancing technology shocks and employment. The overwhelming majority of the literature on these issues is based on data from the US and other developed countries and the empirical literature that uses data from the developing countries is very limited. However, the impact of technology shocks on employment is just as relevant for developing countries, if not more so, in light of the fact that technological progress is widely viewed as a key ingredient of longrun economic growth. The evidence from Korea suggests that while technology shocks contribute to higher productivity in the short run, they may reduce employment in the long run. Such evidence is consistent with the majority of empirical studies on the technology-employment relationship in developed countries.

We hope that our analysis will also contribute meaningfully to the very limited literature on the broader issue of the empirical validity of RBC theory in developing countries, and inspire researchers to pursue the same topic with data from other developing countries in the future. At a broader level, such studies will help developing-country policymakers better understand the forces behind the business cycles of their respective countries and thus provide useful policy guidance. In wake of the current global financial and economic crisis, understanding the sources of macroeconomic fluctuations has become more significant for developing countries. For example, in the case of Korea, our evidence implies that policymakers should pay greater attention to the role of price rigidity as a source of business cycles. In any case, future studies on the impact of productivity-enhancing technology shocks on employment in other developing countries will shed more light on the role of technology shocks on macroeconomic volatility. While technological progress is recognized as an important source of long-run growth in those countries, its impact on short run fluctuations has largely been neglected up to now.

NOTES

1. Examples of standard RBC models include those in Arias, Hansen, and Onanian (2007) and Guvenen (2006). However, Campbell (1994) shows that the effect of technology shocks on hours worked depends on the specific nature of the technological process, and may even have a negative effect.
2. Chang and Hong (2006) and Chang, Hornstein and Sarte (2004) used TFP to investigate the dynamic relationship between technology shocks and employment in US manufacturing sector.
3. See, for example, La Croix (2007), Jorgenson (1995) and Dollar (1991).
4. See Hall (1989) and Mankiw (1989) for more comprehensive discussions.
5. To minimize the misspecification error, Peersman and Straub (2004) used sign restrictions, first suggested by Faust (1998), to identify structural shocks in VAR.
6. Some studies, including McGrattan (2004) and Holzl and Reinstaller (2004) interpreted the two shocks in the structural VAR as technology shocks and demand shocks. However, many supply shocks other than technology shocks, such as shocks arising from fluctuations in production costs or labor supply, have no long-run impact on productivity. Therefore, it seems more appropriate to classify shocks as technology shocks and nontechnology shocks rather than technology shocks and demand shocks.
7. Technology shocks can have a permanent effect on productivity because the level of the TFP is an unstable time series with a unit root.
8. To check for the robustness of our structural VAR results, we also used impulse responses from the standard VAR with Cholesky factorization to construct the innovations. The estimation results are very similar to those from our structural VAR models.
9. To check for the robustness of our structural VAR results, we also used impulse responses from the standard VAR with Cholesky factorization to construct the innovations. The estimation results are very similar to those from our structural VAR models.

Arias, A., Hansen, G., & Ohanian, L. (2007). Why have business cycles become less volatile? Economic Theory, 32, 43-58.
Ball, L., & Moffit, R. (2001). Productivity growth and the Phillips curve. NBER Working Paper No. 8421.
Basu, S., Fernald, J., & Kimball, M. (2006). Are technology improvements contractionary? American Economic Review, 96, 1418-1448.
Basu, S., & Kimball, M. (1997). Cyclical productivity with unobserved input variation. NBER Working Paper No. 5915.
Blanchard, O., & Quah, D. (1989). The dynamic effects of aggregate demand and supply disturbances. American Economic Review, 79(4), 655−673655-673.
Campbell, J. (1994). Interpreting the mechanism: an analytical approach to the stochastic growth model. Journal of Monetary Economics, 33(3), 463−506463-506.
Carmichael, B., Keita, S., & Samson, L. (1999). Liquidity constraints and business cycles in developing economies. Review of Economic Dynamics, 2(2), 370-402.
Chang, Y., & Hong, J. (2006). Do technological improvements in the manufacturing sector raise or lower employment? American Economic Review, 96(1), 352-368.
Chang, Y., Hornstein, A., & Sarte, P. (2006). Understanding how employment responds to productivity shocks in a model with inventories. Federal Reserve Bank of Richmond Working Paper 06-06.
Christiano, L., Eichenbaum, M., & Vigtusson, R. (2003). What happens after a technology shock? NBER Working Paper No. 9819.
Chyi, Y. (1998). Business cycles between developed and developing economies. Review of International Economics, 6(1), 90-104.
Dedola, L., & Neri, S. (2004). What does a technology shock do? A VAR analysis with model-based sign restrictions. CEPR Discussion Paper No. 4537.
Dollar, D. (1991). Convergence of South Korean productivity on West German levels, 1966-1978. World Development, 19(2-3).
Faust, J. (1998). The robustness of identified VAR conclusions about money. Board of Governors of the Federal Reserve System (US), International Finance Discussion Paper 610.
Francis, N., & Ramey, Valerie A. (2005). Is the technology-driven real business cycle hypothesis dead? Shocks and aggregate fluctuations revisited. Journal of Monetary Economics, 52, 1379-1399.
Francis, N., Owyang, M., & Theodorou, A. (2003). The use of long-run restrictions for the identification of technology shocks. Federal Reserve Bank of St. Louis Review, 85, 53-66.
Gali, J. (1999). Technology, employment and the business cycle: Do technology shocks explain aggregate fluctuations? American Economic Review, 89, 249-271.
Gali, J. (2004). On the role of technology shocks as a source of business cycles: some new evidence. Journal of the European Economic Association, 2, 372-380.
Gali, J., & Rabanal, P. (2004). Technology shocks and aggregate fluctuations: How well does the RBC model fit postwar US data? NBER Working Paper No. 10636.
Guvenen, F. (2006). Reconciling conflicting evidence on the elasticity of intertemporal substitution: a macroeconomic perspective. Journal of Monetary Economics, 53(7), 1451-1472.
Hall, R. (1989). Invariance properties of solow’s productivity residual. NBER Working Paper No. 3034.
Holzl, W., & Reinstaller, A. (2004). The impact of technology and demand shocks on structural dynamics: evidence from Austrian manufacturing. MERIT-Influomics Research Memorandum Series 2004-015.
Jorgenson, D. (1995). Productivity. International comparisons of economic growth (Vol. 2). Cambridge, Massachusetts: MIT Press.
Kiley, M. (1998). Labor productivity in US manufacturing: does sectoral comovement reflect technology shocks? Unpublished manuscript, Board of Governors of the US Federal Reserve System.

Kim, S., & Lim, H. (2004). Does Solow residual for Korea reflect pure technology shocks? Presented at the Far Eastern Meeting of the Econometric Society (FEMES) in Seoul, Korea, July 1, 2004.
La Croix, S. (2007). Decomposing and analyzing Korea’s declining GDP growth: Some cautions and suggestions. Department of Economics, University of Hawaii at Manoa, Working Paper No. 7-21, August.
Mankiw, N. (1989). Real business cycles: a new Keynesian perspective. Journal of Economic Perspectives, 3, 79-90.
Masih, A., & Masih, R. (1995). Temporal causality and the dynamic interactions among macroeconomic activity within a multivariate cointegrated system: evidence from Singapore and Korea. Weltwirtschaftliches Archiv, 131(2), 265-285.
McGrattan, E. (2004). Comments on Gali and Rabanal’s 'technology shocks and aggregate fluctuations: how well does the RBC model fit postwar US data? Federal Reserve Bank of Minneapolis, Staff Report 338.

Mendoza, E., & Smith, K. (2006). Quantitative implications of a debtdeflation theory of sudden stops and asset prices. Journal of International Economics, 70(1), 82-114.
Park, H. (2000). Trade and real business cycles in a small open economy: the case of Korea. Bank of Korea Economic Papers, 3(2), 135-163.
Peersman, G., & Straub, R. (2004). Technology shocks and robust sign restrictions in a Euro area structural VAR. European Central Bank Working Paper No. 373.
Shea, J. (1999). What do technology shocks do?. In NBER macroeconomics annual. Cambridge: MIT Press.
Uhlig, H. (2004). Do technology shocks lead to a fall in total hours worked?. Journal of the European Economic Association, 2(2-3), 361−371361-371.
Yoon, J. (2006). International real business cycle model with human capital and skill acquisition activities. Journal of the Korean Economy, 7(2), 253-276.

APPENDIX

See Figure A1, Tables A1 and A2.

Figure A1. Growth of the residual and cyclically adjusted TFP.

Table A1. Unit root tests of the key variables

Notes: Test regressions contain a constant and a linear time trend, and lags of the dependent variable are chosen by AIC, SC and BIC. The null hypothesis is the existence of unit root for ADF and PP tests, and the non-existence of unit root for KPSS test.

• Rejection of the null hypothesis at the 10%10 \% significance level.

** Rejection of the null hypothesis at the 5%5 \% significance level.
∗∗∗{ }^{* * *} Rejection of the null hypothesis at the 1%1 \% significance level.

Table A2. Johansen’s log likelihood test for cointegration

Note: Test regression includes a constant and a linear deterministic trend in the data. The test indicates zero cointegrating equation at the 5%5 \% significance level for each set of the variables.

Available online at www.sciencedirect.com
ScienceDirect