Reallocation effects of recessions and financial crises: an industry-level analysis

3.1 Incidence, duration and amplitude of recessions

From 1970 to 2002, we observe 120 recessions for the 37 country sample as reported in Table 1. The Table reports the sample period for each country (column 2), the cumulative sum of the drop in GDP (column 3) and the mean GDP drop (column 4) during all recessions faced by each country, and column 5, 6, and 7 display the number of recessions, their average duration, and the number of deep recessions, respectively. Seventy-one of those recessions took place within the DV group of countries and the remaining 49 were faced by the EM markets, implying a similar number of recessions per country for both groups. However, sample periods are generally shorter for EM markets, which implies a slightly higher incidence of recessions for that group. Iran underwent the largest number of recessions, 11, between 1970 and 2002 and this clearly places it first in the sum drop of output list. Indonesia experienced the largest average fall in GDP during recessions, but it only experienced one recession in 1997. Other countries like the UK and the US faced five recessions each during the time period considered, with the impact on GDP growth being larger for the UK than for the US. Overall, we can see that the severity of recessions in EM markets exceeds that of DV countries, which is a common feature analyzed in, for instance, Aguiar and Gopinath (2007). This happens not because of a higher incidence of recessions, but because, primarily, recessions in the EM world are deeper. We can also see this by looking into the incidence of deep recessions. 32 out of the 120 recessions were classed as “deep” when considering all countries; six of them took place in DV countries and the remaining 26 in the EM markets. In other words, those 32 episodes produced a higher drop in output than the mean drop of output faced by all countries (2.73%). When using DV and EM country averages as reference groups, we see that for DV countries 29 out of 71 recessions were considered deep, whereas 20 out of 49 recessions are deep for EM economies.

The average duration of recessions is very close for both groups of countries, only slightly shorter for the DV group. On average recessions last about 1 year and 4 months. However, it is likely that this figure is inflated because we only have annual data, setting a floor of 1 year to the minimum recession duration. Finland is the country facing the largest average duration due to the deep and long-lasting depression during the early 1990s. On average, also, recessions tend to happen every 9 years, although this number is slightly shorter for EM countries.

3.2 Industry-specific versus sector-wide recessions

A relevant feature to analyze in the data is whether recession episodes tend to be more concentrated on a few sectors or they are sector-wide events. Note that, given that we identify recessions using GDP and our UNIDO data only contains manufacturing, this may tend to underestimate the incidence of recessions with a sector-specific bias. Nevertheless, comparisons between countries are still possible. Using our definition of recessions, we identify which industries are facing a drop in VA growth in recession years. This enables us to show the average percentage of industries in recession during the episode. That is, whether recessions are coordinated phenomena across industries.^[9]

Another important metric is the standard deviation of the growth rate of VA across industries within a country, which measures the dispersion of VA growth across industries during recession episodes, hence the degree of heterogeneity of performance across industries.

Table 2 shows the average percentage of industries facing negative VA growth during recession years (t=REC) for each country and group. It also shows the percentage of recessions for each country where different percentages of industry branches showed negative VA growth. This enables us to identify whether countries face predominantly industry-specific or industry-wide recessions. We can see that the average percentage of contracting industries at the time of recession episodes is slightly higher for DV than EM countries, 67.43% and 63.98% respectively. This difference however is found to be statistically insignificant. More precisely, from the DV countries considered in this study, Canada and West Germany display the highest percentage of contracting industries at the time of the episode (85.89% and 81.48%, respectively), whereas Ireland displays the lowest percentage out of all the countries (35.19%). From the EM countries group, we can see that in Colombia, Honk Kong, and Indonesia, 88.71%, 84.62% and 86.36% of industries, respectively, are contracting at t=REC. Malta, India and Jordan represent the other extreme in this group.

Perhaps more informative is the second part of the table from which we can see that, in DV countries, 47.14% of the recessions were associated with VA contraction for 70% or more industries, 21.43% with between 60 and 70%, 12.86% with between 50 and 60% of the industries and, finally, only 18.571% with less than 50% contracting industries. In contrast, the numbers for EM markets are consistently lower for high percentages of industries. In fact, almost 37% of recessions were accompanied by less than 50% of industrial branches contracting. Importantly, there is some evidence that the ratio of the variances between DV and EM countries is different to one for 2 out of the 6 grouping classifications, namely, for those groups where 30–40% and 50–60% of industries are in recession at the same time that the aggregate economy is in recession. For the second group the differences in mean between DV and EM countries were found to be strongly significant while for the first one this conclusion only applies if we use a very lax criterion such as a 15% significance level. Therefore, there seems to be some evidence, albeit limited, that recessions tend to be more coordinated across manufacturing industries in DV countries.

Figure 1 shows the average standard deviation of the VA growth rates together with the upper and lower quartile for each group of countries. These graphs are consistent with the results in Table 2, that is, the dispersion of industrial growth rates for EM markets is always higher than for DV economies and of an order of magnitude of almost twice. These results are strongly significant throughout the whole sample (REC–3 to REC+3) and on a year-to-year comparison. Those two sets of results (Figure 1 and Table 2) show the behavior of this metric around recession points. We can see that, while the standard deviation for DV countries increases during recessions (and the year before),^[10] for EM markets the dispersion of growth rates actually increases during the recovery period.

Figure 1:

Standard deviation of VA growth across industries.

These results point to a marked difference between the behavior of sectors across the two groups of countries: while EM markets display more dispersion in VA growth rates, this dispersion behaves counter-cyclically for DV countries and pro-cyclically for EM markets.

3.3 Peaks, troughs, and concordance

We now characterize recessions using a turning points methodology that allows us to unveil the degree of comovement of industries not only for recession episodes, but for all the different stages of the business cycle. We identify turning points in industry cycles following Harding (2002), which is an annual variant of the quarterly Harding and Pagan (2002) algorithm. We apply the algorithm on the log levels of industrial VA. For a given series y_t, a peak (trough) is identified at time t if y_t is higher (lower) than the observations in the preceding and the following year. In particular, a peak is identified in a time series {yt}t=1T at time t if y_t=max{y_t–1, y_t, y_t+1} and a trough is identified at time t if y_t=min{y_t–1, y_t, y_t+1}. We ensure that the peaks and troughs follow pre-specified “censoring rules,” which require: (a) peaks and troughs to alternate; and, (b) the minimum duration of the phase to be 1 year and the minimum duration of the cycle to be 2 years.

Following Harding and Pagan (2002), we then obtain a measure of comovement, known as the concordance index. This index measures the fraction of time two series are in the same phase of the cycle. We make use of this index in three different ways. First, we measure the concordance of phases between an industry and the aggregate business cycle of a given country. This is measured by:

Where S_i,_x,_t and S_x,_t are binary variables capturing expansion and contraction phases of industry i of country x and the aggregate of country x, respectively. Secondly, we measure the concordance of phases between two industries of the same country by:

where i, j are combinations of industries where i≠j and n is the total number of pairwise combinations of industries i and j. Therefore, S_i,_x,_t and S_j,_x,_t are binary variables capturing expansion and contraction phases of industries i and j of country x, respectively.

Finally, we measure the concordance of phases between the same industry, i, across two DV countries or two EM, respectively. The index is estimated across all possible country pairs, n:

where x, y are combinations of industries where x≠y. Therefore, S_i,_x,_t and S_i,_y,_t are binary variables capturing expansion and contraction phases of industry i of country x and y, respectively.

Table 3 summarizes the key characteristics of business cycles for DV and EM countries in terms of duration and amplitude using aggregate GDP data. In terms of duration, differences between the two groups are small. EM countries tend to have slightly longer contractions and expansions, whereas cycles are mildly more asymmetric for DV countries (i.e. differences in the duration of contractions and expansions are larger). The main difference arise in the amplitude of the cycles. As commented above, both contractions and expansions tend to be much larger for EM countries, leading to much higher business cycle volatility.

Table 4 presents the concordance index for industries within countries. The results capture the percentage of time an industry is in the same phase of the cycle as the aggregate or another industry. The first set of columns estimates the index between industries and the aggregate as in equation (1), and the second set estimates the pairwise index as in equation (2). Both sets of results are then aggregated for presentation. In DV countries, on average, a randomly selected industry will be in the same phase as the aggregate economy 61.4% of the time. This number is very similar for EM economies (63.2%). The same conclusion can be obtained when looking at the average pairwise concordance indexes. There are no substantial differences between the two groups of countries. Interestingly, the countries showing the highest degree of concordance are South Korea and the US.

Table 5 presents the results of concordance index as in equation (3) looking at comovement across the same industry between countries. These are concordance indexes for all the pairwise combinations of the same industry across DV and EM countries. They capture the percentage of time two same industries in two different DV or EM countries are in the same cyclical phase. The results are then grouped by industry. They suggest that same industries across DV countries are more often in the same phase than industries across EM countries. This is especially the case for industries that are heavily used for intermediate inputs such as chemical products. While differences are of a small order of magnitude, they are highly statistically significant, and they point out that stronger inter-industry trade linkages may be driving these results.^[11]

4.1 Sectoral activity and shares

Our descriptive analysis will now focus on how economic activity at a sectoral level behaves around recession episodes. We are particularly interested on the evolution of VA, employment, productivity, and VA and employment shares as indicators of sectoral reallocation. Given the definition of a recession discussed above, we plot the evolution of these variables for the 7 years that span the 3 pre-recession and the 3 post-recession years (REC–3 to REC+3). The plots contain the average behavior of the variable across all recessions for each country. We analyzed the results for each country and industry. However, to facilitate presentation, we only report averages for the two groups of DV and EM countries.

Furthermore, because presentation and interpretation is obscured by the large number of industries and variables available, we also collapse industries in four groups depending on their (labor) productivity level. This is because a question of interest, rather than the specific branches themselves, is whether activity reallocates between branches with different productivity characteristics. We classify industries into the following four categories: High, Medium-High, Medium-Low, and Low productivity. We used two different methodologies for this classification. The first simply ranks industries for each country (within) in terms of their productivity levels and assigns them into their corresponding groups by quartiles. The second methodology, rather than using a within country criterion, ranks industries by their level of productivity relative to the same industry in the US. That is, this classification normalizes by the standard dispersion in productivity that exists across different industries because of technical characteristics using the US as the reference country. Although there are some non-negligible differences between these two classification methods regarding the composition of branches, both gave similar results in terms of their behavior around recession points. For this reason, we report here only the results using the first method. Also, this classification is perhaps more interesting as it ranks industries according to their within country productivity level and is hence compatible with a definition of comparative advantage.^[12] All variables were then averaged out for the industries in each group for both groups of DV and EM countries.

Moreover, Rajan and Zingales (1998) identify the level of an industry’s dependence on external finance (the difference between investments and cash generated by operations) from data on US firms.^[13] We make use of this index and collapse industries in four groups depending on the level of EFD: Low to No EFD, Medium-Low, Medium-High, and High EFD. Given that their index provides a measure of an industry’s EFD in the 1980s, we assume that the same ordering will hold for the specific time period under examination in this study, 1970–2002. Importantly, we want to observe whether the depth of the recession and the speed of recovery change for different levels of financial dependency, and whether this result is different between DV and developing countries. All variables were then averaged out for the industries in each group for both groups of DV and EM countries.

Finally, we distinguish between normal and financial recessions by externally identifying banking crises using Reinhart and Rogoff (2008), and compare those episodes between DV and EM economies. From the 120 recessions analyzed in this study, 29 were identified as financial recessions, among which 19 took place in the DV economies and the remaining 10 occurred in EM markets. Appendix B shows the countries and years for which financial recessions took place.

We now present the results grouping industries firstly by levels of productivity and, secondly, by levels of EFD. We then distinguish, on the one hand, between DV and EM economies, and on the other hand, between normal and financial recessions.^[14]

4.1.1 Developed versus emerging economies

4.1.1.1 By levels of productivity for each group of countries

Figure 2 shows the evolution of averaged VA growth from REC–3 to REC+3 for DV and EM countries. Both groups of countries display a V shaped pattern at the REC point. The amplitude of the cycle is larger for EM markets for all groups of productivity levels. Note that, at t=REC, the lower the productivity level of an industry in DV countries the higher the contraction it will face. When comparing REC–3 to REC+3 values, we can see that neither EM nor DV economies recover to pre-recession rates within the 3 years following the recession. Despite that fact, some notable differences exist. EM markets generally face larger contractions than the DV countries, except for the medium-high productivity level group. Moreover, when comparing pre- to post-recession growth rates, while the two highly productive groups of industries face the largest drops in VA growth in DV countries, in the EM economies it is the two lowest productive groups that seem to be affected the most by recession episodes in terms of recovery. The differences observed between DV and EM are strongly significant at the 1% significance level, except for the high productivity group which suggests that the average VA growth in EM countries will be higher than the one in DV with a significance level of 10%.

Figure 2:

Average VA growth per level of productivity for developed and emerging countries.

Similarly, Figure 3 displays the evolution of averaged employment growth from REC–3 to REC+3 for DV and EM countries and results point to strongly significant differences between DV and EM markets at the 5% significance level for all groups of productivity levels. Both groups of countries display a V shaped pattern around the recession time period, although the amplitude of the cycle is larger for DV countries. Moreover, the recovery in employment is much stronger for EM than for DV economies, suggesting a higher degree of real wage flexibility in EM economies.^[15] For the majority of the groups, the deepest contraction is observed the year of the recession. However, notable exceptions exist. On the one hand, the high productivity group of the EM markets does not face negative growth throughout the 7 years of analysis and on the other hand, the low productivity group of the DV countries displays negative growth from REC–3 to REC+3. Importantly, when comparing pre- to post-recession values, this figure shows that on average, the majority of manufacturing sectors in DV countries face very persistent employment losses after a recession. The opposite is true for the EM markets, as for any given productivity level, industries do on average recover to higher growth rates after the recession episode. Interestingly, while the two lowest productivity groups of the EM markets face the largest contractions in VA growth, they face the largest expansions in employment growth, although the latter are bigger than the former. This means that post-recession productivity growth has fallen, which is in line with the results from Figure 4.

Figure 3:

Average employment growth per level of productivity for developed and emerging countries.

Figure 4:

Average productivity growth per level of productivity for developed and emerging countries.

Overall, in DV countries by REC+3 productivity growth has returned to its pre-recession rates. In contrast, for EM markets by REC+3 productivity growth remains below its pre-recession growth rates for all productivity groups except for medium-high, which is also the only group for which results are found to be statistically significant at the 10% significance level.

We also look at the relative dispersion between the high and low productivity groups and the two middle productive groups for VA, Employment and Productivity growth.^[16] Several results stand out. The relative dispersion is higher between high and low productivity groups especially for DV countries. For the EM countries, the relative dispersion is high at REC–3 and REC+3 and falls substantially during the 2 years window around the recession period. For the medium groups, fluctuations are overall around zero. For employment growth, the relative dispersion of the high and low productive groups increases during REC periods for DV countries and falls substantially during the recovery years. For EM this difference shows a constant trend from REC–3 to REC+1, when it then faces a substantial drop in REC+2 before increasing back to its previous levels. Similarly to VA growth dispersion, the medium groups for both DV and EM markets fluctuate around zero. Finally, for the relative dispersion in productivity growth for the high and low productive groups the dispersion decreases before and during recession episodes for DV countries and increases substantially the three following years. The dispersion in EM countries reaches similar levels to the ones observed for DV 3 years after the recession. Importantly however, the relative dispersion cannot highlight different types of dynamics. For instance, the difference can be negative because one group shrinks and the other grows, or because one shrinks more than the other. Qualitatively, both cases are different, although they may display the same type of dispersion.

Figure 5 shows the evolution of averaged VA shares per level of productivity. Shares in general do not display any marked variation around the recession date. Some underlying trends appear to be dominating, especially for the DV countries. But, overall, in DV countries there seems to be a very slight redistribution of VA shares from the lowest productivity group to the three remaining groups, albeit very small. The EM economies are not characterized by any restructuring in VA shares after a recession episode. A very similar picture arises from the evolution of employment shares (Figure 6). While for the average VA share the differences are strongly significant between DV and EM countries for all groups of productivity, this is not the case for the high and medium-low productivity groups when comparing the average employment shares from REC–3 to REC+3.

Figure 5:

Average VA share per level of productivity for developed and emerging countries.

Figure 6:

Average employment share per level of productivity for developed and emerging countries.

Finally, there is clear relationship between industrial productivity level and the distribution of VA and employment shares for EM countries. In particular, the higher the productivity level of an industry the higher the average level of VA shares and the lower the average level of employment shares. As shown later, this could be a consequence of higher sectoral concentration, or specialization, in EM than in DV markets.

4.1.1.2 By level of EFD

The amplitude of the cycle is larger for EM markets, independently of the level of EFD of the industries (Figure 7). When comparing pre- to post-recession values, overall, EM countries face large and persistent output losses, except for the group of industries that have medium-high EFD. For the DV countries, the only group facing gains in VA growth rates is the Low to No EFD, for which recovery occurs within the 3 years following the recession. The two groups with the highest EFD also face the largest contractions from REC–3 to REC+3 (≈2.4%). Moreover, industries with high dependence on external finance face a larger contraction in VA growth in both DV and EM countries. This result is in line with the ones found in Braun and Larrain (2005). All differences observed in the average evolution of VA growth between DV and EM countries are strongly significant for all groups of EFD.^[17]

Figure 7:

Average VA growth per level of external financial dependence (EFD) for developed and emerging countries.

4.1.2 Normal versus financial recessions

In this part we compare normal and financial recessions. Note that to compare results between the previous and the current analysis, one will have to look at the average evolution for all countries and all recessions. However, results are likely to present slight differences as averages are taken by country and not by the number of recessions or industries within a country. In other words, because we assume that countries in our sample are equally important we do not estimate weighted averages to account for the number of recessions in each industry and each category (productivity level or EFD). For instance, because of missing data one country might have only four industries in each grouping instead of seven, which would be the case for a country that has no missing industries. If we were to perform a weighted average to account for the number of industries in each grouping, we would be assuming that industries in the former country are more “important” than industries in the latter. The same would hold for the number of recessions.

Figure 8 shows the evolution of average VA growth per level of productivity from REC–3 to REC+3 for all countries, when normal or financial recessions occur. For any given group of productivity level, we can see that contractions are larger for the case of financial recessions. Moreover, those types of episodes display a W shaped pattern, as growth at REC+1 is at higher levels than pre-recession, but during the following 2 years growth falls to lower levels. Therefore, when comparing REC–3 to REC+3 values, all industries seem to face losses in VA levels, except from those that have low productivity levels. This is the only group that recovers from financial recessions within 3 years. For normal recessions, the recovery is even slower as none of the four groups displays higher post-recession than pre-recession growth. Therefore, whatever the productivity level when normal recessions occur, industries face losses in VA. This result is also supported by Reinhart and Rogoff (2009b) who found that during crises, EM markets face a sharper fall in real GDP growth but a somewhat faster comeback to growth than advanced economies. Similar results are also presented by Calderón and Fuentes (2010). The observed differences are only weakly statistically significant. For the low productivity group, the significance tests suggests that the average VA growth is higher during normal than financial recessions with a p-value=0.0706.

Figure 8:

Average VA growth per level of productivity, normal versus financial recessions.

Figure 9 presents the results for industries ranked by level of financial dependence. Although results are not statistically significant, some interesting patterns arise. As seen before, the amplitude of the cycle is larger for financial recessions. Overall, industries with high EFD face larger contractions in VA growth the year of the financial crises, with contractions being larger for the case of financial recessions. Interestingly, and somehow puzzling, in industries with high EFD the recovery is faster in the case of financial recessions. This result is perhaps consistent with the evidence presented in the work by Calvo, Izquierdo, and Talvi (2006) on “phoenix miracles,” defined as rapid output recovery from financial crises, accompanied by the absence of credit recovery.

Figure 9:

Average VA growth per level of external financial dependence (EFD), normal versus financial recessions.

4.2 Sectoral concentration/specialization: Gini and HHI indexes

We also examined whether recessions are associated with any significant changes in the degree of concentration of VA and employment. We can interpret this concentration as “specialization” as in Imbs and Wacziarg (2003), that is, whether a significant proportion of output (inputs) in the economy is being produced (used) by a few industries. By looking at VA and employment concentration, we can also infer the dispersion of productivity across industries. Whether recessions are associated with greater or lower specialization, of course, will depend on institutions, availability of credit, labor market frictions, changes in the composition of demand, openness, etc. We make use of two different measures: the Gini coefficient and the Herfindahl-Hirschman index (HHI).

The Gini coefficient uses information on how VA and Employment shares are distributed across the different industries. Employment shares have commonly been used in the empirical literature concerning sectoral specialization as a measure of sector size. However, making use of sectoral VA shares helps generalizing the evidence based on sectoral labor inputs.

A simple expression for the Gini index is based on the covariance between the ranked shares of VA or employment by industry, S_R, and the rank that the industry occupies in the distribution of VA or Employment share, F. This rank takes a value between zero for the lowest VA or Employment share and one for the highest. The Gini index, varying between 0 for lowest and 1 for highest inequality, is then defined as:

where S̅_R is the average VA or Employment share.

The HHI is another indicator of the level of concentration/specialization among industries in a sector used in the industrial organization literature. It is defined as the sum of the squared market shares of each industry branch in the sector. Again, we made use of both VA shares and employment shares to obtain the HHI. A decrease in the HHI indicates a decrease in concentration (more diversification). The expression for HHI is then:

where S_i is the share (of VA or employment) of branch i in the manufacturing sector, and N is the number of branches. The HHI (H) ranges from 1/N to one. If all branches have an equal share, the reciprocal of the index shows the number of industries in the sector. The HHI takes into account the relative size and distribution of the industries in a sector and approaches zero when a sector consists of a large number of industries of relatively equal size. The HHI increases both as the number of industries in the market decreases and as the disparity in size between those industries increases. Because of this dependence on N, and given that countries in our sample have unequal numbers of branches, we prefer to use the normalized HHI:

While the H ranges from 1/N to 1, H* ranges from 0 to 1 regardless of the number of branches considered.

Tables 6 and 7 present the Gini and HHI coefficients for sectoral VA and employment shares for DV and EM economies, respectively. Overall, it is obvious that changes in sectoral specialization/concentration are modest, as magnitude changes are in general relatively small. While the differences obtained in the Gini coefficient are statistically insignificant, the ones in the HHI between DV and EM are strongly statistically significant for both VA and employment shares.^[18]

Some important patterns can be observed. When looking at the Gini coefficient, two main conclusions can be drawn. Firstly, the manufacturing sector of DV countries is less specialized around recessions for both VA and employment, when compared to EM markets. Secondly, for both DV and EM countries, employment shares are in general more unequally distributed than VA shares. Although for EM markets the gap between those two measures is marginally larger the 3 years before the recession, at t=REC and the 3 years following the recession this gap becomes larger for the DV economies. This implies that before the recession, productivity is more concentrated in EM than in DV countries. However, at the recession year and the 3 years that follow, productivity becomes more concentrated in DV than in EM countries. Moreover, when looking at the HHI index, results indicate that, with the exception of a few countries like Singapore, Ecuador and Panama,^[19] all countries display low concentration (HHI<0.1) whether using sectoral employment or VA shares. Furthermore, when using either the sectoral VA or employment shares to estimate the HHI, concentration is significantly higher among EM markets than it is among DV countries. Finally, as for the Gini coefficient, employment shares are in general more unequally distributed than VA shares. Therefore, productivity is more concentrated in EM than in DV countries, as the gap between the two measures (VA and employment shares) is larger for the former group of countries throughout the 7 years of analysis. Although the gap is slightly higher for EM markets, after a recession this closes down much more for EM than for DV countries.

4.3 Accounting for structural change: a shift-share analysis and sector misallocation

Shift-share analysis is a descriptive technique to analyze the sources of productivity growth. First proposed by Maddison (1952), it shows how aggregate growth is mechanically linked to differential growth of labor productivity and the reallocation of labor between industries. It has been widely applied for analyzing the effect of industrial structural change on productivity growth (e.g. Fagerberg 2000; Peneder 2003) and microeconomic evidence on the sources of growth (e.g. Foster, Haltiwanger, and Krizan 2001).

Let us define LP=Labor Productivity, VA=Value Added, L=Labor input, and i=industry index with i=(1, …, N). Then,

Define Si=Li∑iLi as the share of industry i in total employment. Then we have that:

Defining ΔLP=LP₁–LP₀, ΔS=S₁–S₀ and using equation (8), we have:

We can express (9) in growth rate form:

The percentage change in labor productivity between time t=0 and t=1 is hence decomposed into three distinct effects. The first component of eq. (10) is the so-called “static-shift effect” and it measures the impact that changes in the allocation of labor between industries have on productivity growth. It will be positive if the share of high productivity industries increases in total employment by attracting more labor resources at the expense of low productivity industries. The second term in (10) is the so-called “within-shift effect” and it measures the change in productivity that would have prevailed if no change in sectoral shares had taken place between 0 and 1. That is, it measures productivity gains that have occurred only within industries. Finally, the third effect is the so-called “dynamic-shift effect.” It captures interactions between changes in sectoral structure and within productivity effects. This effect will be positive if changes in shares favor those industries where productivity is growing. Thus, the “dynamic-shift effect” reflects whether a country reallocates its labor resources towards industries with fast growing productivity.^[20]

Table 8 summarizes the results obtained from the shift-share analysis. It reports the three effects, namely the within, the static, and the dynamic-shift effects, for normal times, recessions and financial recessions, distinguishing DM and EM.^[21]

Overall, the within-shift effect is positive for both DV and EM countries. This result implies that, on aggregate, reallocations of labor between industries (with different productivity levels) do not play an important effect on overall productivity growth. This effect appears to be dominating the structural components, which is in line with results reported in the literature.^[22] It should be emphasized that, at this level of aggregation, all structural shifts between firms within branches will be included in the within effect. To the extent that little resource shift happens between very different branches, we would then expect the static-shift effect to be of a smaller magnitude. The other two effects, the static- and dynamic-shift effects, are more volatile and can be either positive or negative. One pattern that seems to distinguish DV and EM countries is that the dynamic-shift effect plays a more important role in the latter than in the former during recessions. Interestingly, the sign is reversed for this component for EM during financial crises. Finally, the static effect is negative in DV while positive in EM during recession periods, although the opposite is true during financial crises.

In summary, there does not seem to be a clear pattern between the structural components and the recession episodes. Sector-level reallocation does not seem to be associated with the state of the business cycle but rather with longer run trends. Thus, at this level of disaggregation at least, this contradicts theories predicting that, during recessions, there will be more restructuring (i.e. Hall 1991; Caballero and Hammour 1994). Nevertheless, restructuring at the firm level may still be substantial and reflected on within industry effects.

Another relevant issue is whether, for a given level of reallocation, resources are reallocated towards more productive industries. In order to assess this reallocation effect we implement the OP decomposition. There is another way of decomposing productivity which looks at misallocation. Define the aggregate productivity of an economy as P, which is simply the weighted sum of productivities of industries P_i, weighted by their employment shares S_i. P can thus be decomposed in the following way:

where mean(P_i) is the un-weighted mean productivity and the second term is the OP term: the covariance between productivity and sector share. If this covariance is zero, then aggregate productivity is just the mean productivity of its industries. But if large industries have higher productivity, then labor is allocated to the more productive industries and the covariance is positive. Therefore, this is a measure of misallocation: if the OP term is high then resources are more efficiently allocated, if the OP term is negative or zero, then they are not efficiently allocated. Suppose cov(P_i, S_i)=0.2, then this implies that aggregate productivity is 20% higher than if labor was randomly allocated between industries.

Figure 10 reports results for the OP decomposition. For DV countries the level of misallocation does not vary across different types of recessions. A slight upward trend can be observed from REC–3 to REC–3 both during financial crises and “normal” recessions. This trend is sustained for recession periods reducing the misallocation until 1 year after the recession occurs. In the following periods the covariance term remains practically constant. The same is observed for the case of financial crises but at a lower rate. Turning to EM economies, financial crises seem to lead to stronger misallocation 1–3 years after its occurrence. For normal recessions the pattern is very similar to the one observed for DV countries, although the trend is slightly more pronounced.

Figure 10:

Olley-Pakes decomposition between DV and EM countries for normal times, recessions and financial crisis.

In summary, it appears that recessions do not lead to an improved allocation of resources between industries in the manufacturing sector. In EM, in fact, inter-industry misallocation significantly increases during episodes of financial crises.