Selected Issues


Selected Issues

Constraints to Firm Investment and Growth—a Life Cycle Analysis1

This paper highlights growth over a firm’s life cycle as a key factor behind Mexico’s weak productivity growth. It confirms that the average Mexican firm stagnates after some 10–15 years of age at less than twice its initial size. The few industries with stronger life cycles tend to be located close to the US border and related to the North American supply chain. While NAFTA appears to have boosted life cycle growth only for initial cohorts of firms in some sectors (e.g. transportation), this does not appear to be the case across the board. The analysis also uncovers important distortions that explain weak life cycle growth. These include the prevalence of informality, highlighting that informal firms do not grow while formal firms experience solid and continuous growth. Moreover, stronger firm growth appears to be associated with less concentrated industries that are less likely to be subject to the undue use of market power. Firms are also less likely to invest and grow if they are far away from population centers and do not have access to good financial and internet services.

A. Motivation

1. Mexico implemented sweeping structural reforms during the mid-1990s following a series of economic and financial crises. The reforms succeeded in achieving macroeconomic stability, opened the economy up to trade and foreign investment, and boosted educational attainment.

2. Against this backdrop, Mexico’s negative productivity growth in recent decades remains a puzzle (Levy and Rodrik, 2017). Previous work (e.g. Levy, 2018; Misch and Saborowski, 2018) has shown that productivity in Mexico is low in part because the allocation of capital and labor is inefficient. A complementary explanation would be that individual firms are simply not productive enough.

3. One explanation for low firm productivity could be insufficient investment due to the prevalence of distortions that discourage firm growth. Hsieh and Klenow (2014) argue that the lack of investment constrains productivity both directly in incumbent firms and by reducing competition for new entrants. In this context, their finding that manufacturing plants in Mexico grow far less than firms in the United States as they age is an important concern.

4. The aim of this paper is to understand why firms in Mexico do not invest more and thus remain relatively small and unproductive. We make use of several waves of the Mexican Economic Census and other subnational data sources to calculate firm growth over firms’ life-cycles. We then decompose life cycle dynamics to understand regularities and attempt to identify distortions that explain the lack of firm growth.

B. Methodology and Data

5. Estimating firm life cycles is challenging. A firm’s life cycle measures the evolution of size over its lifetime. The ideal research setting would thus be one in which the researcher could simply aggregate observed firm-level life cycles for all firms in the economy. In reality, however, longitudinal data on firm growth is typically not available for all types of firms.

6. Firm life cycle dynamics have been calculated using different approaches, each associated with important shortcomings. Two of the more well-known papers on this topic are Hsieh and Klenow (2014) and Eslava and Haltiwanger (2017). Hsieh and Klenow (2014) calculate firm life cycles based on a specific wave of the Mexican Economic Census by calculating the average firm size in a given 6-digit industry for a series of age groups. By combining average firm sizes across age groups into a representative life cycle, the approach faces two main problems: first, it does not follow the same firms over time and, second, it is likely associated with attrition bias since earlier age-groups will include a larger number of unproductive and, crucially, relatively small, firms that do not make it to the older ages.2 Eslava and Haltiwanger (2017) avoid both problems by focusing on longitudinal firm-level data for Colombia, but their data is incomplete in that it does not include the full universe of firms as small and informal firms are underrepresented.

7. We make use of several waves of the Mexican Economic Census to calculate life cycles for Mexican firms. As is standard in the literature, we measure firm size as the number of employees. We use the same data source as Hsieh and Klenow (2014) but makes use of a larger number of waves of the Economic Census to ensure that we can follow the same firms over time. Specifically, we define age groups in five-year intervals (0–4 years, 5–9, 10–14, 15–19, 20–24). We then compare the average firm size in age group 0–4 in industry i and state s in the 1993 wave to the average size of firms in age-group 5–10 in the same industry and state in the 1998 wave. This gives us the average growth rate of firms in industry i and state s from age group 0–4 to age group 5–10. We continue this process until the 2013 wave which gives us the full life cycle of firms in industry i and state s born in the 1993 wave from ages 0–4 to ages 20–24. We label these firms the 1993 cohort.

8. We calculate life cycles for four cohorts of firms per industry and state. We follow the same process as described in the previous paragraph for the cohorts of firms born in the 1998, 2003 and 2008 waves, except that the life cycles we calculate for these firms are progressively shorter. For example, we can draw out the evolution of firms in the Bakery and Tortillas sector in Aguascalientes that were 0–4 years of age in 1993 until they were 20–24 years old. But we can only observe the cohort of firms in the same industry and state that was 0–4 years old in 1998 up to age 15–19. Throughout our analysis, we include only cohorts that we can observe at birth. So, for instance, we do not include the cohort of firms that was 5–9 years old in 1993. This leaves four cohorts per industry-state pair whose life-cycles we observe. Table 1 illustrates the structure of the resulting data set for better exposition. Note that firm sizes are normalized to one at birth to allow better aggregating the data across states and sectors.

Table 1.

Mexico: Stylized Example of Data Set

article image

9. Our approach improves upon the methodology used by Hsieh and Klenow (2014) and covers a broader sample of firms than Eslava and Haltiwanger (2017). While we are, like Hsieh and Klenow (2014), not able to work with longitudinal firm level data, we do follow the same cohort of firms over time. We also make an effort to directly control for the problem that cohorts may get smaller over time because small and unproductive firms exit disproportionately (attrition bias).3 In contrast to Eslava and Haltiwanger (2017), in turn, our data set includes the full universe of non- agricultural firms in Mexico, including small and informal ones. Moreover, we distinguish life cycles not only at the level of the 6-digit industry but also by state, allowing us to run regressions that include state and industry as well as wave fixed effects.

C. Descriptive Analysis

10. Mexican firms no longer grow after some 10–15 years of age. We focus on the 1993 cohort of firms to illustrate life cycle dynamics in Mexico. When we aggregate the life cycles of all industry state pairs across states, and then across industries by taking medians, we find that the median sector in Mexico sees firms less than double in size before stagnating after 10–15 years of age (Figure 1, left chart). The picture looks even more bleak when we correct for a potential attrition bias (Figure 1, right chart). We see somewhat more and more continues growth in the manufacturing sector, but the sector does not look much different when correcting for attrition bias. This finding is broadly in line with Hsieh and Klenow (2014) who estimate that manufacturing firms in Mexico stagnate after reaching around two times their initial size around age 20. The authors find that US firms, by contrast, continue growing throughout their lifetimes to more than seven times their initial size.

Figure 1.
Figure 1.

Mexico: Aggregate Life Cycle Dynamics Across Industries

Citation: IMF Staff Country Reports 2018, 308; 10.5089/9781484383711.002.A002

11. Only a small share of firms experiences significant growth in higher ages. Figure 2 illustrates the distribution of industry life cycles. As can be seen only a very small share of Mexican firms grows to high multiples of their initial size, with even the 90th percentile remaining far below Hsieh and Klenow’s (2014) results for the US economy.

Figure 2.
Figure 2.

Mexico: Distribution of Life Cycles Across Industries

Citation: IMF Staff Country Reports 2018, 308; 10.5089/9781484383711.002.A002

12. Firms with stronger life cycles appear to be clustered along the US border and in industries related to the North American supply chain. A closer look at the manufacturing sector suggests that many of the sectors related to the North American supply chain are among those that experience the highest growth over their life cycles (e.g., transportation, food and basic metals). However, there also appears to be notable regional variation suggesting that, beyond industry composition there may be an important role for structural rigidities in determining life cycles. Figure 3 illustrates that life-cycles in the manufacturing sector are weakest in Mexico’s less developed South and stronger closer to the border with the US.

Figure 3.
Figure 3.

Mexico: Median Manufacturing Life Cycle Growth by State1

Citation: IMF Staff Country Reports 2018, 308; 10.5089/9781484383711.002.A002

1/ The map shows the growth the average firm in the state experiences over its life cycle. Specifically, the underlying figures reflect the median across all 6-digit industries in the state.

13. The benefits from NAFTA appear to have been restricted to earlier cohorts of firms in some industries. An interesting question is whether cohorts that were the first to benefit from NAFTA (e.g., the 1993 and 1998 cohorts), some of which broke into new markets for Mexican firms, experienced stronger life cycles than those who came later. Looking at the transportation sector, the answer appears to be that initial cohorts indeed experienced more growth over their lifetimes than later cohorts (Figure 4). For example, firms in the 1993 cohort in the motor manufacturing sector grew in size by a factor of 44 by the time they were 5–10 years of age while the 1998 and 2008 cohorts experienced 10 and 1.3-fold growth, respectively. A potential explanation might be that NAFTA had a sort of level effect on firm growth in that firms initially grew fast until global car producers were saturated in the share of production they would be interested in bringing to Mexico. An alternative explanation is that firms in later cohorts came in with larger initial investments than the initial cohorts.

Figure 4.
Figure 4.

Mexico: Life Cycles in the Transportation Sector

Citation: IMF Staff Country Reports 2018, 308; 10.5089/9781484383711.002.A002

14. However, this finding cannot be generalized to all fast-growing manufacturing sectors. The basic metals sector is a case in point in which life cycles weakened over time in some industries while they strengthened in others (Figure 5). This may suggest that, in contrast to a sector like the transportation sector that is dominated by a relatively small number of large firms, in other sectors the benefits from NAFTA may have survived.

Figure 5.
Figure 5.

Mexico: Life Cycles in Basic Metals Sector

Citation: IMF Staff Country Reports 2018, 308; 10.5089/9781484383711.002.A002

D. Econometric Analysis

15. In this section, we aim to better understand the determinants of firm growth over the life cycle in Mexico. Our econometric specification takes the form


where Qi,s,t,c is the growth the average firm experienced between birth and wave t in industry i and state s in cohort c (e.g. the cohort that was 0–4 years old in 1993 would be c=1993). All regressions include fixed effects for each 6-digit industry i, state s and wave t. They further include age group dummies DAG for each industry-state pair which are intended to capture average life-cycle dynamics. The set of fundamentals Xi,s,t can vary by industry, state and wave (year). The latter are computed based both on data from the Mexican Economic Census and the SIMBAD database which includes a range of socioeconomic indicators with variation at the municipal level (which, by matching these indicators with locational information for firms in the Economic Census, allows creating regressors that do not drop from the regression even in the presence of industry and state fixed effects). Finally, we also allow for interactions DAGXi,s,t between age group dummies and the fundamentals to allow computing average life cycles conditional on fundamentals.

16. The regressions make use of up to 56,800 observations. All regressions include the three sets of fixed effects which are omitted in the tables. Regression 1 in Table 2 includes only the age group dummies in addition to the fixed effects in the regression. With life cycles normalized to 1 in the initial age group, the coefficients suggest that the average industry sees firms grow to some 2.4 (1 +1.4) times their initial size after 20–24 years of age once fixed effects are controlled for. While this suggests a somewhat stronger average life-cycle than in the descriptive results, the results are closer to the initial findings (at 1.5–1.6 times the initial size) once outliers (the largest and smallest 1 percent realizations of the dependent variable) are dropped in Regression 2. Importantly, the R² rises from 0.07 to 0.22 once these outliers are dropped.

17. We control both for the initial average firm size as well as for a potential attrition bias in the regressions. Regression 3 additionally includes two control variables to correct for potential biases. The initial average size of firms in the sector is meant to even the playing field between industries with initially very large and industries with initially very small firms. As can be expected, it appears that life cycle growth is somewhat weaker when firms are large from the outset. The variable measuring the share of the initial cohort that survived until the given age group is our control for potential attrition bias. As expected, it appears that firm growth is higher in sectors in which a smaller share of the initial population of firms survives as it is likely small and unproductive firms that do not make it to the higher ages.

Table 2.

Mexico: Regression Table

article image
Robust standard errors in brackets*** p<0.01, ** p<0.05, * p<0.1

18. Informality is a key determinant of the strength of firm life cycles. Regression 4 includes all those fundamentals that we are able to observe in all years associated with the waves in the sample. Additional variables that we are only able to compute for later years are included in Regressions 5–7 as they would otherwise reduce the number of observations in the baseline regressions. A crucial determinant of life cycles appears to be the degree of informality in a given sector-state pair. Informality here is calculated as the share of firms that pays neither social security contributions nor VAT (and is thus unlikely to be registered at either tax authority). It appears that industries with a higher prevalence of informality experience sharply weaker life-cycles. This result is unsurprising since informal firms are not only less likely to invest and grow themselves (e.g. to remain under the radar of tax authorities or due to lack of access to bank credit) but also may depress investment in more productive formal firms that cannot compete due to the unfair cost advantages enjoyed by informal firms.

19. Informal firms grow only marginally over their lifetimes while formal firms experience solid and continuous growth. When including not only the informality term itself but also its interaction with the age group dummies DAGXi,s,t in the regression (not shown in the table), we can draw a set of life cycles as predicted by the regression for fully informal and fully formal industry state pairs. As shown in Figure 6 (top left chart), informal firms on average hardly grow at all after the age of 4 years over their life cycles when controlling for other explanatory variables. Formal firms, in contrast, not only do not stagnate but grow to some 4.5 times their size by the age of 20–25.

Figure 6.
Figure 6.

Mexico: Predicted Life Cycles Conditional on Regressors

Citation: IMF Staff Country Reports 2018, 308; 10.5089/9781484383711.002.A002

20. Market concentration also weakens firm growth, likely due to the undue use of market power that discourages investment in the majority of firms. Using interaction terms between the age group dummies and the concentration variable, we find evidence of a close association between life cycle growth and market concentration (defined as the Herfindahl index calculated based on the 20 largest firms in the sector state pair), suggesting that firms invest and grow more in markets that are less concentrated and thus perhaps less subject to the undue use of market power. The top right chart in Figure 6 shows that firms in very concentrated industries (99th percentile of the distribution of the Herfindahl index) have notably weaker life cycles than those in very diversified industries (1st percentile of the distribution).

21. Another interesting relationship arises between life cycle growth and the average geographical distance of firms from major markets (defined as the distance from the closest city with more than 500,000 inhabitants). It appears that more remotely located firms do not experience as much growth as firms closer to not only customers but also competitors and suppliers. Once again using interaction terms in the regressions, the middle left chart in Figure 6 shows how much stronger life cycles are in municipalities located very closely to large cities as opposed to municipalities located very far from large cities.

22. Firms also appear to have stronger life cycles when they have access to financial and internet services. Intuitively, financial and internet services may discourage investment into firm growth. The middle right and bottom left charts in Figure 6 show how life cycles in industries with full coverage of financial and internet services, respectively, compare to those with zero coverage according to the predictions of our baseline regression augmented by interaction terms.

E. Conclusions

23. This note finds that firm life cycle growth in Mexico is significantly lower than in more advanced economies such as the US. The underlying driver, namely the failure to invest, may explain part of the productivity slowdown the Mexican economy has experienced in recent decades. In particular, the average Mexican firm appears to no longer grow after some 15 years of age and to stagnate at less than double its initial size.

24. We also find that several of the few industries with stronger life cycles are those located in Mexico’s North, close to the US border. While NAFTA appears to have delivered only a temporary boost to life cycles in some sectors (e.g., transportation), this does not appear to be the case across the board.

25. The analysis also uncovered important distortions that appear to be associated with weaker firm investment and growth. These include the prevalence of informality, highlighting that informal firms do not appear to grow much over their life cycles while formal firms experience solid and continuous growth. Moreover, stronger firm growth appears to be associated with less concentrated industries that are presumably less likely to be subject to the undue use of market power. Firms are also less likely to invest and grow if they are far away from population centers and do not have access to good financial and internet services.

26. These findings highlight the importance of pushing ahead with the important structural reform agenda begun with the Pacto por Mexico in 2012. This includes the fight against informality, strengthening competition in the Mexican economy as well as access to financial and telecommunication services. Furthermore, targeted infrastructure investments could help better connect the more remote regions to the major markets of the country.


  • Haltiwanger, J., and M. Eslava, 2017, “The Drivers of Life-cycle Growth of Manufacturing Plants,” 2017 Meeting Papers 1540.

  • Hsieh, C.-T., and P. J. Klenow, 2014, “The Life Cycle of Plants in India and Mexico,” The Quarterly Journal of Economics, 129(3), pp. 103584.

    • Search Google Scholar
    • Export Citation
  • Levy, S., 2018, “Under-Rewarded Efforts: The Elusive Quest for Prosperity in Mexico, Inter-American Development Bank.

  • Levy, S., and S. Rodrik, 2017, “The Mexican Paradox,” Available at

  • Misch, F., and C. Saborowski, 2018, “Resource Misallocation and Productivity: Evidence from Mexico”, IMF Working Paper, 18/112.


Prepared by Florian Misch (FAD) and Christian Saborowski (WHD).


In a slight variation to this approach, the authors calculate life cycles based on two waves of the Mexican Economic Census by comparing the average size of a given age group in the first wave to the average size of five-year older firms in the same sector in the second wave. This somewhat attenuates the first shortcoming but does not alleviate the selection bias.


We do so by running a regression of life-cycle levels on the share of firms in a cohort that survived until a given point in the life cycle. We then calculate the corrected life cycle as error term in the regression.

Mexico: Selected Issues
Author: International Monetary Fund. Western Hemisphere Dept.