Demand Shock and Productivity: What Determines Firms’ Investment and Exit Decisions?∗ Pradeep Kumar†, University of Exeter Hongsong Zhang‡, University of Hong Kong March 10, 2016 Preliminary Abstract What determines firms’ decisions on capital investment and exit? The industrial organization literature encompasses that they are driven by productivity alongside other deterministic factors, which determine firms’ long-term expectation of profits. In this paper, we investigate the roles played by the short-term demand shocks—besides productivity—on firms’ capital investment and exit decisions. The use of inventory data allows us to infer the magnitude of the short-term demand shocks separately from supply-side productivity shock. We find that while both productivity and demand shocks are important factors in determining firms’ investment decisions, demand shock is the major driving force for a firm’s exit. These findings have important implication for understanding firms’ behavior. Keywords: Demand Shock, Productivity, Inventory, Production Function Estimation, Investment, Firm Turnover.

∗

We are grateful to Mark Roberts, James Tybout, Uli Doraszelski, Paul Grieco and the participants of the IO reading workshop at Penn State University for very insightful comments and suggestions. We also thank Mark Roberts and James Tybout for providing data for this research. All errors are our own responsibility. † Pradeep Kumar: Department of Economics, University of Exeter Business School, UK. [email protected]. ‡ Hongsong Zhang: Faculty of Business and Economics, The University of Hong Kong, Hong Kong. [email protected].

1

1

Introduction

What determines intra-industry firm turnover and firm-level capital investment decisions? In the field of industrial organization and international trade, the non-deterministic component is usually attributed to productivity, which is a long-term firm characteristic and is persistent over time. Based on the theoretical foundation by Jovanovic (1982) and Hopenhayn (1992) , seminal papers of Olley and Pakes (1996) in industrial organization and Melitz (2003) in international trade, study firm exit behavior driven by a supply-side productivity measure. Most empirical literature that followed, using either firm-level or plant-level data, have found evidence to support this idea using a revenue-based productivity measure. In practice, however, firms face substantial amount of transitory demand shocks in their daily operation. These shocks may come from different sources, such as transitory preference shocks, income shocks or an unexpected change in weather. The transitory demand shocks may have an impact on firm turnover and investment decisions, especially when firms faces credit constraints and/or when manager/shareholders’ sentiment is affected by these shocks, which could be quite common in practice. In this paper, we explore the roles played by the shortterm demand shocks, besides the persistent productivity, on firm turnover and investment decisions. A major challenge is to separate the transitory demand shocks from productivity, when neither is observed in the data. A solution proposed in the literature is to rely on output price data to model the demand side explicitly whenever the output price is available, as in Foster et al. (2008) and Roberts et al. (2013). These approaches, however, are not feasible if the price data is not observable to the researcher, which is quite common in most production datasets. In this paper, we propose an alternative way to use inventory stock at the firm level to disentangle productivity and demand-side shocks. Our model is based on the premise that the inventory stock contains important information about demand shocks. We assume that each firm targets a fixed level of inventory. This allows us to attribute the variation in the inventory stock to the hidden demand shock. One advantage of our method, compared with the aforementioned price-based approach, is that inventory information is usually recorded in many production datasets, such as Colombia plant level survey and Chinese manufacturing survey. This makes our method widely applicable. Methodologically, our model extends the popularly used Olley and Pakes (1996) approach to explicitly allow for the inventory level to affect the firms’ input decisions. This extension provides one way to address the multi-collinearity problem prevailing in the Olley and Pakes 2

style models,1 because the inventory stocks and demand shock provide independent variation between firms’ investment decisions and static labor and material decisions. We estimate our model using a plant-level dataset from Colombia, which has detailed information on plant-specific inventory stocks. Estimation results from three representative industries (including Clothing, Plastic Products, and Leather Shoes) show that both productivity and demand shocks are important characteristics of a firm. The inter-quartile range for productivity ranges from 6.83% to 25.48% in the three industries after controlling for demand shocks and inventory, which is much smaller than that estimated without controlling for demand shocks and inventory. The inter-quartile range for demand shocks also ranges from 8.13% to 11.83% in these three industries according to the estimation results, implying that the 75th percentile firm has a demand shock that is 8.13% to 11.83% higher than the 25th percentile firm. We also find large heterogeneity of both productivity and demand shocks for firms of different types. In particular, new entrants on average have lowest productivity and medium demand shocks, and exiting firms have lowest demand shocks and medium productivity. In contrast, continuing incumbent firms have both highest productivity and highest demand shocks among the three groups. We also find that firms become more productive as they get older, and they tend to get higher demand shocks as well. Presumably, these results suggest a learning of both the market conditions and the customer base (demand side), as well as a learning of the production process (supply side) at the firm level. These empirical evidence further supports that the demand shock should be given ample consideration in firm behavior analysis. We estimate the firms’ exit decisions and investment decision as implied by the theoretical model, using two different methods. Both methods provide similar estimates. For all industries, demand shocks have a negative and significant effect on firm exit rate. Increasing demand shock by one percent reduces exit probability by about 1%. In contrast, the impact of productivity on exit rate is insignificant after controlling for demand shock. This suggests that firm exit is mainly driven by the short-term demand shock which presumably causes financial or other problems to force firms out of the market. On this point, our findings are consistent with Foster et al. (2008) and Roberts et al. (2013). For the investment decisions, we find a positive and significant effect of demand shocks on investment in all three industries. The productivity effect is positive in all three industries, but not significant in 1

As Ackerberg et al. (2006) and Bond and Soderbom (2005) pointed out, there is a multi-collinearity problem in the Olley and Pakes (1996) first stage estimation, because both the investment and labor choice are functions of the same variables: capital, productivity and age. To estimate the labor/material coefficient consistently in the first stage, we need an independent variation between the labor and investment.

3

Plastic Products industry. The magnitude of the effect from demand shocks is larger in the industries where both productivity and demand shock have significant effects on investment. This result again suggests that the short-term demand shock is an important firm characteristic determining firms’ investment decisions. This paper is related to two other line of study. First, it is related to a large literature on the determinants of capital investment. A survey by Chirinko (1993) notes that the vast majority of literature on determinants of investment finds it to be a function of prices, output levels and shocks. The author pointed out, “relatively little work has been done on quantifying the effect of autonomous shocks on investment”. There have been some studies addressing this gap since then, most of which use aggregate data. For example, Ghosal and Loungani (1996) finds a negative relationship between investment and price uncertainty (which could result from demand shocks and/or productivity shocks) at the industry level. More recently, Bloom et al. (2007) find that higher uncertainty reduces the responsiveness of investment to demand shocks. Audretsch and Elston (2002) support the role of demand factors in providing more liquidity in the investment behavior of German firms. Cooper and Ejarque (2003) study the role of financial frictions in firm investment behavior using a dynamic optimization model. Second, this paper relates to the research on determinants of firm turnover, which has been a long standing field of study. These studies have found market frictions, demand-learning and market size as important factors driving firm exit. A survey study by Tybout (2000) finds that high turnover in manufacturing firms doesn’t necessarily imply less productive firms are driven away. He notes that market frictions are important in determining firm turnover in developing economies. Dixit and Chintagunta (2007) find that both supply and demand factors are responsible for firm exit in the airline industry. The authors put particular emphasis on the effect of firms’ learning behavior on market exit but abstract away from the role of supply-side productivity. Similarly, Disney et al. (2003) study the role of learning in firm exit in the U.K. manufacturing industry. Asplund and Nocke (2006) study the role of market size as determinant of firm exit. More recently, Collard-Wexler (2013) finds smoothing of demand fluctuations has a significant impact on firm exit decision using a dynamic oligopoly model. The rest of the paper is organized as follows. Section 2 constructs an econometric model to separate demand shocks from productivity at the firm level. Section 3 reports the estimation results and examine the basic features of the recovered demand shocks and productivity. Section 4 analyses the roles of demand shocks and productivity on firm behavior. Section 4

5 check the robustness of our results to alternative specifications. Finally, we conclude in Section 6.

2

The Model

We develop a dynamic model of firm production by incorporating demand shocks and inventory stock into the standard framework of Olley and Pakes (1996). The extended model allows us to recover productivity and demand shocks at the firm level from observed revenue data which are typically available in many widely used firm/plant level datasets. Even though we do not observe price levels in the data, we are able to quantify the demand shock using inventory stock data.

2.1

The Model Setup

The production function is assumed to be Cobb-Douglas, αk αl Yjt = exp(ωjt + jt )Kjt Ljt Mjtαm ,

(1)

where Yjt , Kjt , Ljt and Mjt represent the output level, capital stock, labor and material stocks respectively. The parameters αk , αl and αm are the associated factor share parameters. Firm j has a productivity level of ωjt , which is structural. The production is subject to a non-structural productivity shock jt . Following Doraszelski and Jaumandreu (2013) and Aw et al. (2011), the productivity is assumed to follow a first order Markov process, ωjt = g(ωjt−1 ) + ηjt ,

(2)

where ηjt is the current period innovation in the productivity. We assume that ηjt is i.i.d. across firms and over time. We assume that productivity innovation, ηjt , is realized at the beginning of each period. Firms observe their own capital stock (kjt ), productivity (wjt ) and last period inventory stock (invjt−1 ) in the beginning of a period and choose labor (ljt ) and material (mjt ) to produce output (yjt ) without knowing the exact market condition. After production, demand shock (zjt ) is realized which determines the sales and hence leads to the realization of inventory stock (invjt ) and period profits (πjt ). Finally, firms choose whether to exit and their investment (ijt ) levels. 5

The timing assumption that production decisions happen before the realization of demand shock captures the fact that firms usually do not have complete information about the market demand of their product when production happens (though they may have some expectation). The demand shock, zjt , represents the uncertainty firm j faces at time t when it determines how much to produce. The firm’s ex-ante optimal choice of output level may not be ideal after the realization of demand shock, in the sense that it may generate too much or too little inventory. As a result, the level of inventory shock contains information about the demand shock. We want to use the information contained in the inventory stock to recover the level of a firm’s demand shock. The demand shock, although i.i.d. drawn, has a dynamic effect on firms’ future production and profitability, like productivity. However, the mechanism is different. The demand shock has its dynamic effect by changing the firms’ end-of-year inventory levels, which affects firms’ investment and exit decision. The production decision is dynamic as well since it affects the end-of-period inventory levels, which is a determining factor in next period’s production decision. As a result, all three choices made by a firm (production, exit, and investment) are dynamic when demand uncertainty and inventories are introduced into the model. The firms’ decisions can be summarized by two Bellman equations. The Bellman equation corresponding to the production decision at the beginning of each time period is, V1 (kjt , wjt , invjt−1 ) = M ax Ezjt [πjt + V2 (kjt , wjt , invjt )]. ljt ,mjt

(3)

The choice of labor and material is based upon the capital stock, productivity shock and last period’s inventory stock. Note that the labor and material decisions are dynamic since their choice determines the inventory stock, which is a state variable affecting firms’ future choices. The Bellman equation corresponding to the investment (ijt ) and exit (xjt ) decisions is, V2 (kjt , wjt , zjt , invjt ) = M ax Eηjt [βV1 (kjt+1 , wjt+1 , invjt )] ijt ,xjt

subject to:

(4)

ijt ≤ Rjt

where Rjt summarizes all resource available to firm j at time t. The constraint captures potential financial constraints a firm may face: the level of investment is usually influenced by the firms’ available resource, either through direct cash flow or collateral (see Midrigan and Xu (2014) for example). Demand shock affects firms investment and exit decisions because it directly influences the resources available to a firm. Demand shock also affects 6

investment and exit decisions through a rather indirect channel by driving the inventory levels, a state variable, which affects the firms’ dynamic choices. The introduction of demand shock and inventory stock into the model has multi-fold implications. First, it brings in another dimension of firm heterogeneity which plays an important role in a firm’s daily operation. This allows us to study and compare the role of the two heterogeneities (demand shock and productivity) in a firm’s key economic decisions. Second, the inclusion of demand shock and inventory stock implies that the labor and material choices are dynamic. Third, it naturally suggests one way to recover both productivity and demand risk the firm faces by utilizing the readily available information on inventory. Moreover, the timing assumption that the demand shock realizes before the choice of investment but after labor and material choices provides one possible way of breaking the multi-collinearity in the first stage of standard Olley and Pakes (1996) estimation as criticized by Ackerberg et al. (2006).

2.2

Inventory and Demand shock

The first step in our estimation procedure is to quantify demand shock in a simplified way in order to avoid explicitly solving the dynamic model with multidimensional choices. In each time period, we have the following accounting equation, Yjt + invjt−1 = Sjt + invjt

(5)

where Yjt is the production amount and Sjt is the sales by firm j in time period t. The above feasibility equation notes that the sum of production quantity and beginning-of-period inventories equate to the sum of sales and end-of-period inventory stock. Demand shock, zjt , is defined as the disturbance in the expected sales of a firm, Sjt = exp(zjt )E(Sjt |Ijt )

(6)

where Ijt is the information set of a firm j at time t when it is making its production decision, which includes the beginning of period state variables in the value function V1 . We assume that the i.i.d. demand shock, zjt , follows a mean zero normal distribution with standard deviation σ. The mean zero assumption implies that the sales prediction by a firm will not be biased in either positive or negative direction. In other words, no firm consistently under-predicts or over-predicts the demand. 7

To proceed further, we need to make an assumption about the firm’s inventory choice behavior. Firms may intentionally overshoot in their production and maintain an optimal level of inventory stock because of the uncertainty about the demand level and probably for other reasons like production smoothing in the presence of a volatile productivity shock. Eq. (3) implicitly determines firms’ optimal inventory decision as well, besides labor and material choices. To capture this idea, we assume that each firm j at time t targets an inventory stock, λjt , which is a fixed share of the expected sales E(Sjt |Ijt ). λjt = λj E(Sjt |Ijt )

(7)

This seems to be a reasonable assumption over the short to medium term.2 From a firm’s perspective, when it is making its production decision, the available output must equal the expected sales plus the targeted inventory stock. Hence, the optimal production output of a firm must satisfy the following equation, Yjt + invjt−1 = E(Sjt |Ijt ) + λjt

(8)

It is still an accounting equation in the ex-ante sense, but it also captures firms’ optimal inventory and production decisions. Eq. (5) to (8), which are based on firms’ optimal decisions as depicted in section 2.1, suggest a way to recover the demand shock. More specifically, we can we first insert Eq. (8) into (7) to solve out the expected sales E(Sjt |Ijt ), E(Sjt |Ijt ) =

Yjt + invjt−1 . 1 + λj

Inserting this equation in (6) and using (5), we have  log 1 −

invjt Yjt + invjt−1

 = − log(1 + λj ) + zjt

(9)

This equation links a firm’s realized end-of-year inventory stock to its optimal inventory strategy, λj , which is firm specific, and a demand shock zjt , which is transitory. The firmspecific term associated with the firms’ optimal inventory share, − log(1 + λj ), is identified 2

We make this simplifying assumption as modeling the production smoothing motive of a firm by allowing firms to choose the optimal inventory stock each time period (λjt ) is not the main objective of this paper. Also, allowing inventory choice and demand shocks simultaneously make the model very complicated. The optimal choice of inventory stock in the production function estimation literature is another interesting idea in itself and we leave this for future research.

8

from cross-firm variation of the average inventory invjt . The transitory demand shock, zjt , is identified by the within-firm variation of inventory over years. In terms of empirical estimation, we explicitly model firms’ optimal inventory share, λj , as a function of firm characteristics. More specifically, we assume that a firm’s optimal inventory share is a function of its size, ownership, and location, λj = f (Xj ) = f (sizej , ownj , loctj ).

(10)

Replacing λj in (9) by (10),  log 1 −

invjt Yjt + invjt−1



= f˜(sizej , ownj , loctj ) + zjt ,

(11)

where the f˜(·) function represents the term − log(1 + λj ) with λj replaced by (10). We can estimate this equation directly by approximating the f˜(·) function by a suitable polynomial and using data on beginning-of-year inventory, end-of-year inventory, output, firm size, ownership, and location. The demand shock is the residual itself, and the optimal inventory share can be recovered from the regression function. In Section 5, we use an alternative approach, by treating the optimal inventory strategy as a firm fixed effect, to confirm that the estimation results are robust to the parametric form of inventory share in Eq. (10).

2.3

Zero Inventories

When demand is very high, the realized sales will be high and the firm will have zero inventory, as shown in (5). As a result, the inventory choice faces a truncation problem: it is positive when the demand shock is below a critical value, and zero when the demand shock is above this critical value. For instance, around 7% of the observations have zero inventories in our data for the clothing industry. Two issues arise in the estimation of Eq. (9) and (11) from the aforementioned truncation problem. First, using the OLS will bias the estimate of the fixed effects term. In Eq. (9), the dependent variable equals zero when inventory is zero (so that Sjt = Yjt + invjt−1 ) and it is negative when inventory is positive. This problem can be addressed by using a Tobit model. The second issue is more serious. We need the magnitude of demand shock to estimate its prediction power in firm activities (e.g. investment behavior and exit decision). Even if the optimal inventory share λj (or the parameters in f˜(·) in the parametric approach) 9

is consistently estimated, we cannot recover the exact magnitude of demand shock when inventory is zero. For these observations, given the limited information, we replace zjt by its expectation conditional on zero inventory. Under the normal distribution assumption for ˆ j is consistently estimated, the conditional demand shocks and that the inventory share λ expectation of demand shock is defined as follows, Z E(zjt |invjt = 0) =

+∞

¯j λ

φ(zjt ) σ 1 zjt dzjt = √ ¯ ¯ j ) exp 1 − Φ(λj ) 2π 1 − Φ(λ

! ¯2 −λ j . 2σ 2

As a result, the demand shock we constructed for firm j at time t is as follows,

zˆjt =

   log 1 −

1 ¯j ) 2π 1−Φ(λ

  √σ



invjt + Yjt +invjt−1  −λ¯2 

exp

ˆ j ) when invjt > 0, log(1 + λ

j 2σ 2

(12)

when invjt = 0.

This is a simple way to predict the underlying demand shock by using the limited information in a truncated data.3

2.4

Demand Shock, Inventory, and Productivity

After estimating the demand shock, we use a two-stage approach to estimate the productivity by extending Olley and Pakes (1996). We recognize that the demand shock affects firm choices, and as a result production estimation, through the appearance of inventory stock in the model. First stage of the estimation starts with the following production equation in logarithm, yjt = βl ljt + βm mjt + βk kjt + wjt + jt , where jt can be interpreted as a measurement error or the part of the productivity unobserved to the firm. It is well known in the literature that running this regression equation directly suffers from endogeneity issues. To deal with this, we use the insight of ? and control for productivity using firms’ energy expenditure, ejt , which is assumed to be monotonic in productivity. Using energy expenditure instead of investment levels, commonly used in the literature, is particularly important when we want to compare the driving force of supply-side productivity and the demand shock in the determination of investment levels. 3

We can also use a simulation-based approach, which is very similar to the above conditional expectationbased approach. After estimating Eq. (9) or (11) using Tobit, we can derive the conditional distribution of zjt for observations with zero inventory. Then we can take multiple sets of random draws from this estimated distribution to quantify the average effect of demand shocks on firm activities.

10

Using energy expenditure instead of investment levels to back-out supply-side productivity increases the confidence that our results are not forced by the structure of the model. Replacing the demand shock zjt by its estimates zˆjt , we can recover productivity, ωjt , from the energy policy function, wjt = wt (ejt , kjt , zˆjt , invjt ). The underlying assumption in the inversion of the policy function is the monotonicity between energy expenditure and productivity. Energy expenditure is more easily adjustable than investment levels which is usually lumpy. This suggests a stronger monotonic relationship using the energy expenditure. Our approach differs from OP in that we control for the demand shock and its resulting inventory stock when recovering the unobserved productivity. The first stage estimation equation can be re-written as, yjt = βl ljt + βm mjt + βk kjt + wt (ejt , kjt , zˆjt , invjt ) + jt = βl ljt + βm mjt + φ(ejt , kjt , zˆjt , invjt ) + jt If we assume labor and material to be static as is the case in OP, βl and βm can be estimated consistently in the first stage only. But labor and material are dynamic variables due to the introduction of savings in our model. Ackerberg et al. (2006) argue that the first stage of OP suffers from multi-collinearity problems in this case since their is no independent variation between labor/material and investment. This collinearity problem becomes even more severe in Levinsohn and Petrin (2003), which uses intermediate inputs instead of investment to recover the unobserved productivity in the first stage. Introduction of the demand shock provides a way of solving this collinearity problem in the spirit of Ackerberg et al. (2006). The i.i.d. demand shock affects the investment but not the labor and material choices, and as a result overcomes the collinearity problem by providing an independent variation between labor/material and φ(ejt , kjt , zˆjt , invjt ).4 This independent variation gives us identification for βl and βm even when labor and material choices have dynamic implications. From the first stage, φˆjt = yˆjt − βˆl ljt − βˆm mjt . To estimate the capital coefficient, βk , we use the Markov assumption on productivity evolution process, φˆjt = βk kjt + g(φˆjt−1 − βk kjt−1 ) + ηjt 4

Ackerberg et al. (2006) also suggested using a new timing to identify the model, or to estimate the labor coefficient together with all other coefficients in the second stage to avoid the collinearity problem.

11

The last equation forms the basis of the second stage estimation and the capital coefficient, βk , is consistently estimated from it. The productivity measure can then be constructed by wjt = φˆjt − βˆk kjt .

3

(13)

Estimation Results

This section discusses data and the estimation of productivity and demand shock.

3.1

Data and Summary Statistics

The data used in this paper is from the Colombian manufacturing census from 1977 to 1991, which was collected by the Departamento Administrativo Nacional de Estadistica (DANE). It contains detailed information about plants’ domestic and imported inputs usage, output, and many other plant characteristics. We estimate the model for three industries: clothing, plastics and leather shoes. We choose these three varied industries since they are important ones for the economy, have significant inventory shares (greater than 20%) and have sufficient observations (greater than 3500). For a detailed introduction to the data, please refer to Roberts and Tybout (1996). Table 1 shows the summary statistics for each of the three industries. Inventory share is calculated as the ratio of value of inventory to the value of sales at the firm level. There are two things to note here. First, inventory accounts for a large share in firms’ sales. The inventory-to-sales ratio in the three industries range from 22% to 28%. Second, given an industry, the variation of inventory share across firms is substantial. The ratio of mean to standard deviation of the inventory share across industries ranges from 0.59 to 0.75. This variation in the inventory stock across firms and time in an industry suggests that the demand shock can have a large dispersion as well. Also, inventories reported are point sampled at the end of the year. This data limitation would have been problematic if we had chosen an industry that produces perishable goods such as food products, since cannot be stored for a longer term. The summary statistics for exit rate, capital investment, labor expenditure, capital stock, material expenditure, age, and sales are also reported in Table 1.

12

3.2

Production Function and Productivity Evolution

The estimation results for output elasticity for each input and productivity evolution process are reported in Table 2. Capital output elasticity is 0.1% for clothing industry, 0.09% for plastic products industry, and 0.07% for leather shoes industry. Labor output elasticity is highest in clothing industry at 0.38% and roughly 0.3% in the other two industries. Material output elasticity is 0.37% in clothing industry and roughly 0.57% in the other two industries.5 Productivity evolution is fairly persistent in all three industries (0.6-0.7). In general, all three industries show decreasing returns to scale.

3.3

Productivity and Demand Shock

A key output from production function estimation is the implied productivity distribution of firms within an industry. Given the parameter estimates, the productivity can be recovered from Eq. (13). Their summary statistics are reported in Table 3. It is shown that there is substantial amount of dispersion for both productivity and demand shocks among firms within each industry. For productivity (in logarithm), the inter-quartile range is 0.2548 in the clothing industry, implying that the 75% quantile firm is roughly 25% more productive than the 25% quantile firm. This number is slightly lower in the other two industries, with 0.1257 for plastic products, and 0.0683 for leather shoes. The ten-to-ninety percent quantile range is 0.4978 in the clothing industry, implying that the productivity for the ninety percentile firm is about 1.5 times as high as that for the ten percentile firm. This number is 0.2526 for plastic products and 0.1372 for leather shoes. A second key output of our estimation is the distribution of demand shock within a industry. Given the parameter estimates, a measure for demand shock can be calculated from Eq. (12). The dispersion of demand shock is also substantial within an industry, although its dispersion is smaller than that for productivity. The inter-quartile range is between 0.0813 to 0.1183 in these three industries, and the ten-to-ninety percent quantile is between 0.1924 in plastic products industry to 0.2801 in clothing industry. Both of these facts suggest a substantial heterogeneity in demand shocks across firms. Table 4 compares average productivity and demand shocks across entrants, incumbents and exiting firms. In almost all cases, incumbents have a higher productivity and demand shock than entrants and exiting firms. Exiting firms have lower demand shocks than entrants and 5

Note that output elasticities are same as the Cobb-Douglas shares since the model is logged.

13

incumbents on an average for the three industries. Although, exiting firms have a lower productivity than incumbents in the clothing and plastics industry, but the order reverses in the leather shoes industry. This suggests the importance of demand shocks in driving firm turnover. Also, entrants have the lowest average productivity in the three categories. Table 5 further reports the correlation pattern between the two recovered measures, productivity and demand shock, and firms’ input and output indicators. It is shown that productivity is positively correlated with firms’ input choices and sales. The correlation is especially high for labor and material choices. In contrast, the correlation between demand shock and input choices are close to zero, as assumed in the model. However, the demand shock is positively correlated with sales, with a average correlation of about 0.1 in the three industries.

3.4

Inventory Share

We also recover firms’ inventory share, λj , as a side product while recovering demand shock. The estimates are reported in Table 6. On average, the estimated inventory to expected-sales ratio is about 10% for the clothing industry, with a standard deviation of 0.0306 showing a reasonable amount of dispersion across firms within one industry. The estimated inventory to expected-sales ratio is about 7.5% for plastic products and leather shoes industries, with a standard deviation of 0.0221 and 0.0448 respectively. These estimates are in the reasonable range as observed in the data.

4

Productivity, Demand Shocks, and Firm Behaviour

It is of fundamental importance to understand what drives a firm to exit, and what factors affect its investment. In this section, we use our estimation results to explore the connection between firm heterogeneity (productivity and demand shock) and firm behavior dynamics (exit and capital investment decisions). The major purpose behind this exercise is to determine the relative importance of technology versus demand factors in driving the turnover and growth of firms. The analysis also acts as a robustness check for the productivity and demand shock measures constructed in the earlier sections.

14

4.1

What Drives Firms to Exit?

An important application of the productivity measures in the existing literature is to understand the firm turnover in operation. Table 1 shows the average exit rate of firms in each industry. The exit rate is defined as the share of firms that stopped operating to the total incumbents in each year. For example, in the clothing industry 14.67% firms exit at the end of each year on an average. Firm exit rate in the other two industries are of similar magnitude, suggesting that it is a common feature in these industries. We want to examine what drives a firm to exit. In particular, we want to separate roles played by the supply-side shock (productivity) and market-side shock (demand shock) in determining a firm’s exit decision. Our model in Eq. (4) implies that firms’ decision to exit depends on productivity, demand shock, capital size, and inventory stock. As a result, we estimate a Probit model of firms’ exit decisions based on the following equation e xjt = x(ωjt , zjt , kjt , invjt , Xjt ) + ξjt ,

(14)

e is an i.i.d. shock to a firm’s where Xjt are other control variables such as firm age, and ξjt

exit decisions, which is assumed to be uncorrelated with firm states. We report the detailed regression results in table 7, 8 and 9. The dependent variable equals 1 if a firm exits during a year, otherwise it equals 0. The first two columns report results from isolated regressions. We find that a higher demand shock reduces the probability of exiting significantly. In contrast, although productivity also has a negative coefficient, it is not significant in two of the three industries. In column 3 and 4, we control for age and firm size (capital stock). Again, we find similar effect of demand shock on firm exit: a good demand shock significantly reduces the probability of exiting. In contrast, the effect of productivity on exit remains insignificant in two industries. In column 4, we observe that firms that are larger and older are less likely to exit even after accounting for the shocks, consistent with the findings in Olley and Pakes (1996), Dunne et al. (1988) and Dunne et al. (1989). In our preferred full regression in column 5, we further control for inventory size. We find that the effect of demand shock remains statistically significant and in fact becomes even stronger. In the three industries overall, an increase of demand elasticity by one percent reduces the probability of firm exit by 0.8%-1%. After controlling for demand shock and firm size, firms with larger inventory stock are less likely to exit—although higher inventory implies lower demand shock. However, the effect of productivity on firms’ exit

15

decisions is insignificant in all three industries. Overall, the estimation results imply that the demand shocks play a more important role in determining a firm’s exit decision, while productivity may not be as important. This finding has a substantial implication: a firm’s exit may not be mainly driven by its persistent productivity; instead, it is more likely to be forced out due to a low demand shock. It implies that the firm turnover analysis based on productivity alone (excluding demand shocks) conducted in the literature can be misleading for certain industries—especially industries with a volatile demand. Hence, to forecast a firm’s exit decision more reliably, we want to stress the need to consider another more important aspect of firm heterogeneity, demand shock, besides productivity.

4.2

What Drives Firms to Invest More?

Productivity measures are often used to understand firm growth, via capital investment for example. In this subsection, we test the roles played by productivity and demand shocks in shaping firms’ growth via capital investment. Note that our model in Eq. (4) implies that firms’ capital investment decision is a function of its productivity, demand shock, capital size, and inventory stock. Accordingly, we estimate a model of firms’ investment decisions based on the following equation i ijt = i(ωjt , zjt , kjt , invjt , Xjt ) + ξjt ,

(15)

i where ξjt is an iid shock which is independent of firm states. We use TOBIT as our pre-

ferred model to estimate the investment decision because investment is usually lumpy with a substantial amount of zeros in the data. As a robustness check, we also estimate the investment decision using OLS. In the first two columns of Tables 8, 9 and 10, we estimate the effect of productivity and demand shock on investment separately, after accounting for firm age, size, and inventory. It turns out that both productivity and demand shock have an positive impact on investment, except that the productivity effect is insignificant in the plastic industry. Column three of Tables 8, 9 and 10 reports the estimation results from the full model as captured in Eq. (15). Again, demand shock in all three industries has a positive and significant impact on investment. In contrast, productivity affects investment positively and significantly in two of the industries we investigated, and in the other industry (plastics) it is effect is positive 16

but not significant. Moreover, the estimation also shows that the effect of demand shock on investment is higher that of productivity in all three industries, even when the productivity effect is significant. This implies that in general, the short-term demand shock is a more important factor affecting firms’ investment decisions, compared with the more persistent productivity which has been thought as the major determinant of firm investment decisions traditionally. Also, the TOBIT estimation suggests that firms that are younger, larger, and with a higher level of inventory tend to investment more. To check the robustness of these results, we also estimate the investment function Eq. (15) using OLS, and the results are very similar.

5

Robustness Checks

In our primary estimation, we assume that firms’ inventory share is a parametric function of firm characteristics including firm size, ownership, and location, as specified in (10).6 We also use an alternative approach to estimate the inventory share and demand shock, without adding any parametric assumptions on firms’ inventory share. Instead, we estimate Eq. (9) by treating the inventory share as a firm-specific fixed effect. The obvious advantage of this method that we leave the firm-level inventory share, λj , completely flexible and is guided by data only. The limitation of this approach, however, is that it requires a long panel data to estimate the firm effect with credibility. The panel data used has 15 periods, which is arguably long enough to estimate Eq. (9) with firm dummies. However, the panel is unbalanced with a much shorter tenure on average for each firm. In practice, we keep firms which are present for 4 years or more in the data. This leaves us a smaller subsample for each industry, with 8,405 observations for clothing industry, 2,600 for plastic products, and 2,208 for leather shoes industry. After estimating demand shock using Eq. (9), we also estimate the productivity using the method outlined in Section 2.4. Then we use these alternative measures of demand shock and productivity to estimate regression equations (14) and (15). These results are reported in the Appendix. The estimation results are similar to our primary method results. In particular, the estimation results support that the short-term demand shock is the major driving force in firm exit decisions, but not the persistent productivity. Also, both productivity and demand shocks have a positive effect on firm investment. Although some of these effects are not significant statistically because of the smaller sample size, the signs of the parameters are 6

For empirical implementation, we used a third degree polynomial function with interactions.

17

same as earlier.

6

Conclusion

We examine the roles played by productivity and demand shocks on firms’ exit decisions and capital investment decisions. We use inventory data to infer the magnitude of the unexpected short-term demand shocks at the firm level and separate it from productivity. This is different from the existing literature which uses output price data to quantify the demand shock. Since inventory data is more easily available in production datasets, our method is more easily applicable. Then, we use our estimates to investigate the separate roles played by the short-term demand shocks and persistent productivity on firms’ capital stock investment and exit decisions. We find that the demand shock is an important driving force that explains firms’ exit decisions, whereas the effect of productivity vanishes once we control for the demand shock. Also, demand shocks always have a positive and significant effect on firms’ investment decisions in all three industries we examined. In contrast, although the effect of productivity on capital investment is positive in all industries in question, it is insignificant in one of the industries after controlling for demand shocks and inventory. These results imply that the role of short-term demand shocks, as one source of firm heterogeneity, is at least as important as productivity in determining firm exit and investment behavior, if not more.

References Ackerberg, D. A., K. Caves, and G. Frazer (2006): “Structural Identification of Production Function,” Working paper, UCLA Economics Department. Asplund, M. and V. Nocke (2006): “Firm Turnover in Imperfectly Competitive Markets,” Review of Economic Studies, 73, 295–327. Audretsch, D. and J. Elston (2002): “Does firm size matter? Evidence on the impact of liquidity constraints on firm investment behavior in Germany,” International Journal of Industrial Organization, 20, 1–17. Aw, B. Y., M. Roberts, and D. Y. Xu (2011): “R&D Investment, Exporting, and Productivity Dynamics,” American Economic Review, 101, 1312–1344. Bloom, N., S. Bond, and J. V. Reenen (2007): “Uncertainty and Investment Dynamics,” Review of Economic Studies, 74, 391–415. 18

Bond, S. and M. Soderbom (2005): “Adjustment Costs and the Identification of CobbDouglas Production Functions,” Unpublished Manuscript, The Institute for Fiscal Studies, Working Paper Series No. 05/4. Chirinko, R. (1993): “Businss Fixed Investment Spending: Modeling Strategies, Empirical Results, and Policy Implications,” Jouranal of Economic Literature, 31, 1875–1911. Collard-Wexler, A. (2013): “Demand Fluctuations in the Ready-Mix Concrete Industry,” Econometrica, 81, 1003–1037. Cooper, R. and J. Ejarque (2003): “Financial frictions and investment: requiem in Q,” Review of Economic Dynamics, 6, 710–728. Disney, R., J. Haskel, and Y. Heden (2003): “Entry, Exit and Establishment Survival in UK Manufacturing,” The Journal of Industrial Economics, 51, 91–112. Dixit, A. and P. K. Chintagunta (2007): “Learning and Exit Behavior of New Entrant Discount Airlines from City-Pair Markets,” Journal of Marketing, 71, 150–168. Doraszelski, U. and J. Jaumandreu (2013): “R&D and Productivity: Estimating Endogenous Productivity,” Review of Economic Studies, 80, 1338 – 1383. Dunne, T., M. Roberts, and L. Samuelson (1988): “Patterns of Firm Entry and Exit in U.S. Manufacturing Industries,” The Rand Journal of Economics, 19, 495–515. ——— (1989): “The Growth and Failure of U.S. Manufacturing Plants,” The Quarterly Journal of Economics, 104, 671–698. Foster, L., J. Haltiwanger, and C. Syverson (2008): “Reallocation, Firm Turnover, and Efficiency: Selection on Productivity or Profitability?” American Economic Review, 98, 394–425. Ghosal, V. and P. Loungani (1996): “Product Market Competition and the Impact of Price Uncertainty on Investment: Some Evidence From US Manufacturing Industries,” The Journal of Industrial Economics, 44, 217–228. Hopenhayn, H. A. (1992): “Entry, Exit, and firm Dynamics in Long Run Equilibrium,” Econometrica, 60, 1127–1150. Jovanovic, B. (1982): “Selection and the Evolution of Industry,” Ecnometrica, 50, 649– 670. 19

Levinsohn, J. and A. Petrin (2003): “Estimating Production Functions Using Inputs to Control for Unobservables,” The Review of Economic Studies, 70, 317–341. Melitz, M. J. (2003): “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity,” Econometrica, 71, 1695–1725. Midrigan, V. and D. Y. Xu (2014): “Finance and Misallocation: Evidence from Plantlevel Data,” American Economic Review, 104, 422–458. Olley, G. S. and A. Pakes (1996): “The Dynamics of Productivity in the Telecommunications Equipment Industry,” Econometrica, 64, 1263–1297. Roberts, M., D. Y. Xu, X. Fan, and S. Zhang (2013): “A Structural Model of Demand, Cost, and Export Market Selection for Chinese Footwear Producers,” NBER Working Paper 17725. Roberts, M. J. and J. R. Tybout (1996): “Industrial Evolution in Developing Countries: Micro Patterns of Turnover, Productivity, and Market Structure,” Oxford ; New York : Published for the World Bank by Oxford University Press. Tybout, J. R. (2000): “Manufacturing Firms in Developing Countries: How Well Do They Do, and Why?” Journal of Economic Literature, 38, 11–44.

Appendices A

List of Tables

Table 1: Summary Statistics Industry Obs Clothing 12,649 Plastics

4,164

Leather Shoes

3,571

Age 10.65 (9.11) 12.31 (9.59) 11.37 (10.10)

Inv. Inv. Sh. 13.18 0.2750 (3.08) (0.36) 14.27 0.2429 (2.32) (0.41) 12.86 0.2244 (2.89) (0.30) 20

Exit% Invest. Labor 14.67% 2.95 14.13 (35.38) (10.37) (1.10) 13.95% 4.58 14.59 (34.65) (11.89) (1.27) 15.59% 4.11 13.97 (36.28) (9.84) (1.21)

Capital 13.12 (1.43) 14.64 (1.78) 13.21 (1.60)

Mat. Sales 14.70 15.51 (1.51) (1.16) 15.68 16.36 (1.64) (1.48) 14.75 15.42 (1.36) (1.25)

Table 2: Production Function Parameters Industry Capital Share Labor Share Material Share Clothing 0.1042*** 0.3847*** 0.3674*** (0.0019) (0.0047) (0.0032) 0.2924*** 0.5708*** Plastics 0.0927*** (0.0017) (0.0058) (0.0044) 0.3035*** 0.5767*** Leather 0.0681*** (0.0015) (0.0065) (0.0058)

Persistence(ω) Scale 0.7055*** 0.8563 (0.0076) 0.5985*** 0.9559 (0.0132) 0.7184*** 0.9483 (0.0159)

Table 3: Demand shock and Productivity Productivity Industry

Mean

Clothing 3.3449 Plastics 1.8101 Leather 1.7982

Demand Shock

SD

P75/P25

P90/P10 Mean

SD

P75/P25

P90/P10

0.2169 0.1151 0.0802

0.2548 0.1257 0.0683

0.4978 0.2526 0.1372

0.1399 0.1171 0.1098

0.1183 0.0823 0.0813

0.2801 0.1924 0.2164

-0.0068 -0.0028 -0.0037

Table 4: Demand shock and productivity across groups. Productivity

Demand Shock

Industry

Entrants

Incumbents

Exiters

Entrants

Incumbents

Exiters

Clothing

3.293 (0.2100) 1.7897 (0.1306) 1.7987 (0.0791)

3.3632 (0.2137) 1.8151 (0.1064) 1.8028 (0.0758)

3.3246 (0.2243) 1.8014 (0.1165) 1.8040 (0.0965)

-0.0139 (0.1506) -0.0036 (0.1355) -0.0018 (0.1133)

-0.00009 (0.1310) 0.0010 (0.1022) -0.0021 (0.1036)

-0.0259 (0.1575) -0.0086 (0.1317) -0.0132 (0.1252)

Plastics Leather

Table 5: Correlation patterns Productivity

Demand Shock

Industry

Capital

Labor

Material

Sales

Capital

Labor

Material

Sales

Clothing Plastics Leather

0.2301 0.2133 0.1257

0.6089 0.5197 0.4689

0.4705 0.4881 0.4217

0.6755 0.5469 0.4858

-0.0265 0.001 -0.0073

0.0613 0.0256 0.0161

-0.0038 0.1088 0.0529

0.1144 0.1245 0.0883

Table 6: Descriptive statistics of the inventory share Industry

Mean

Clothing 0.1042 Plastics 0.0757 Leather 0.0776

Std. Dev

Median

P25

P75

0.0306 0.0221 0.0448

0.1091 0.0741 0.0751

0.0786 0.0601 0.0434

0.1299 0.09 0.1055

21

Table 7: Probit Regression for Exit model ω z age

(1)

(2)

(3)

Clothing -0.3015*** -0.2747*** (0.0783) (0.0817) -0.6577*** -0.6233*** (0.1080) (0.1131) 0.0093*** (0.0018)

capital

(4)

(5)

-0.1401* (0.0840) -0.7211*** (0.1140) .0093*** (0.0018) -0.1000*** (0.0123)

-0.1030 (0.0834) -0.8983*** (0.1161) 0.0102*** (0.0018) -0.0740*** (0.0129) -0.0425*** (0.0056) 9,882 0.2978 2779.4

inventory Obs Psuedo R2 LR Chi2 ω

9,882 0.2789 2602.71

10,035 0.2858 2729.08

9,882 0.2850 2659.99 Plastics -0.4970* (0.2919) -0.4606** (0.2328) 0.0115*** (0.0031)

-0.3137 (0.2820) -0.4867** (0.2297)

z age capital

9,882 0.2920 2725.24 -0.3217 (0.2942) -0.5030** (0.2341) 0.0152*** (0.0032) -0.1125*** (0.0182)

inventory Obs. Psuedo Rˆ2 LR Chi2 ω

3198 0.3259 974 -0.2952 (0.3972)

z age capital

3207 0.3252 975.98

3198 0.3316 991.09

3198 0.3445 1029.56

Leather Shoes -0.2613 -0.3587 (0.4174) (0.4279) -0.8298*** -0.7395*** -0.7492*** (0.2555) (0.2657) (0.2674) 0.0073** 0.0111*** (0.0031) (0.0032) -0.1293*** (0.0204)

inventory Obs. Psuedo R2 LR Chi2

2754 0.2734 742.19

2777 0.2747 754.61

2754 0.2786 756.42

22

2754 0.2937 797.32

0.1081 (0.3090) -0.7925*** (0.2402) 0.0158*** (0.0032) -0.0641*** (0.0211) -0.0705*** (0.0153) 3198 0.3514 1050.17 -0.2249 (0.4369) -0.9927*** (0.2770) 0.0117*** (0.0032) -0.0971*** (0.0225) -0.0406*** (0.0115) 2754 0.2981 809.31

Table 8: Investment Regression model

TOBIT (1)

ω z age capital inventory Pseudo R2 Obs LR Chi2/F-stat ω

capital inventory Psuedo R2 Obs LR Chi2/F-stat ω z age capital inventory Pseudo R2 Obs. LR Chi2/F-stat

(3)

(4)

Clothing 1.9489*** 1.3599*** 1.8382*** (0.3529) (0.3636) (0.2308) 4.1616*** 3.6691*** 2.1547*** (0.5263) (0.5454) (0.3452) -0.1025*** -0.0916*** -0.1017*** -0.0623*** (0.0081) (0.0078) (0.0080) (0.0051) 2.9619*** 3.0367*** 2.9685*** 1.9894*** (0.0579) (0.0559) (0.0579) (0.0343) 0.0819*** 0.1229*** 0.1249*** 0.0899*** (0.0258) (0.0250) (0.0266) -0.0169 0.0541 0.0545 0.0548 0.2784 12225 12495 12225 12225 3504.07 3599.56 3549.75 247.83 1.2449 (0.8001)

z age

(2)

OLS

-0.0231** (0.0090) 2.0248*** (0.0598) 0.2823*** (0.0494) 0.0767 4105 1852.66

Plastics 0.4033 (0.8150) 4.0598*** 3.9088*** (0.7447) (0.7576) -0.0219** -0.0220** (0.0089) (0.0089) 1.9595*** 1.9748*** (0.0603) (0.0604) 0.3973*** 0.3639*** (0.0482) (0.0520) 0.0776 0.0778 4122 4105 1881.83 1879.33

Leather Shoes 3.3047* (1.8746) 3.8722*** 3.5675*** (1.3334) (1.3714) -0.1096*** -0.1026*** -0.1068 (0.0146) (0.0142) (0.0146) 2.9585*** 3.0051*** 2.9329*** (0.1093) (0.1077) (0.1097) 0.2739*** 0.2642*** 0.3135*** (0.0594) (0.0583) (0.0615) 0.0632 0.0623 0.0635 3467 3516 3467 1143.02 1144.13 1149.83 4.3340** (1.8323)

23

1.4761** (0.6564) 3.1115*** (0.6123) -0.0172** (0.0073) 1.7362*** (0.0477) 0.2657*** (0.0405) 0.399 4105 142.75 6.2534*** (1.1577) 1.8969** (0.8359) -0.0613*** (0.0090) 1.9619*** (0.0623) 0.1995*** (0.0363) 0.3221 3467 86.19

Table 9: Production Function Estimation: Intercept Method Industry

Capital Share

Clothing 0.0587*** (0.0022) Plastics 0.0580*** (0.0018) Leather 0.0473*** (0.0027)

Labor Share

Material Share

Persistence

Scale

0.4920*** (0.0057) 0.2795*** (0.0068) 0.3431*** (0.0083)

0.3503*** (0.0036) 0.6077*** (0.0049) 0.5440*** (0.0073)

0.7346*** (0.0084) 0.7784*** (0.0131) 0.6857*** (0.0181)

0.901 0.9452 0.9344

Table 10: Demand Shock and Productivity: Intercept Method Productivity Industry

Mean

Clothing 2.6652 Plastics 1.9354 Leather 1.9665

Demand Shock

SD

P75/P25

P90/P10

Mean

SD

P75/P25

P90/P10

0.1574 0.0907 0.0911

0.1772 0.1134 0.0634

0.3413 0.2248 0.15

-0.0023 0.0003 -0.0041

0.0955 0.08 0.0734

0.0575 0.0392 0.0398

0.1727 0.1177 0.1448

Table 11: Descriptive Stats. for inventory share: Intercept Method Industry Desc.

Mean

Std. Dev

Median

P25

P75

Clothing Plastics Leather

0.1061 0.0776 0.0775

0.1217 0.0887 0.1021

0.0753 0.0485 0.0403

0.0220 0.0199 0.0112

0.1477 0.1062 0.1132

24

Table 12: Probit Regression for Exit: Intercept Method model ω

(1) -0.2754* (0.1507)

z age

(2)

(3)

Clothing -0.3419** (0.1559) -0.9891*** -0.8692*** (0.2072) (0.2165) 0.0163*** (0.0026)

capital

(4)

(5)

-0.0060 (0.1604) -1.0186*** (0.2183) 0.0160*** (0.0027) -0.1215*** (0.0168)

-0.2603 (0.1644) -1.1619*** (0.2203) 0.0185*** (0.0027) -0.0819*** (0.0180) -0.0515*** (0.0073) 6,463 0.0511 178.47

inventory Obs Psuedo R2 LR Chi2 ω z age

6,463 0.0070 24.5

6,528 0.0127 44.8

6,463 0.0229 79.79

Plastics -1.8354*** -2.4103*** (0.5368) (0.5659) -1.4585*** -1.5213*** (0.4529) (0.4651) 0.0195*** (0.0055)

capital

6,463 0.0378 131.78 -1.3115** (0.6058) -1.3081*** (0.4643) 0.0207*** (0.0056) -0.1282*** (0.0290)

inventory Obs. Pseudo R2 LR Chi2 ω

1957 0.0351 34.44 -1.4983*** (0.5561)

z age capital

1,959 0.0328 32.27

1957 0.0575 56.45

1957 0.0767 75.30

Leather Shoes -1.4289** -0.7876 (0.5644) (0.6045) -1.8307*** -1.5697*** -1.6546*** (0.4850) (0.4991) (0.5116) 0.0120*** 0.0152*** (0.0044) (0.0046) -0.1665*** (0.0300)

inventory Obs. Pseudo R2 LR Chi2

1,645 0.0232 23.3

1652 0.0292 29.38

1,645 0.0409 41.11

25

1,645 0.0718 72.10

-0.8587 (0.6421) -1.2532*** (0.4674) 0.0207*** (0.0056) -0.1026*** (0.0314) -0.0539** (0.0246) 1957 0.0814 79.98 -0.7665 (0.5859) -1.5141*** (0.5174) 0.0157*** (0.0046) -0.1444*** (0.0328) -0.0269* (0.0153) 1,645 0.0748 75.12

Table 13: Investment Regression for the Intercept Method model

TOBIT OLS (2) (3) (4) Clothing ω 2.5853*** 2.1338*** 2.6571*** (0.5873) (0.6011) (0.3917) z 3.9407*** 3.2356*** 2.0195*** (0.8714) (0.8915) (0.5846) age -0.1315*** -0.1179*** -0.1285*** -0.0801*** (0.0102) (0.0099) (0.0102) (0.0067) capital 2.8805*** 3.0034*** 2.8872*** 1.9739*** (0.0708) (0.0653) (0.0708) (0.0433) 0.1237*** 0.1329*** 0.1368*** 0.1060*** inventory (0.0301) (0.0293) (0.0303) -0.0199 Obs 8,405 8,506 8,405 8,405 Pseudo-R2 / R2 0.058 0.0583 0.0583 0.2961 LR Chi2/F-stat 2620.70 2664.08 2633.96 207.49 Plastics ω 0.1162 0.1079 1.6700 (1.5598) (1.5604) (1.2749) z 0.5261 0.4308 0.3275 (1.1853) (1.1858) (0.9982) -0.0326*** -0.0337*** -0.0326*** -0.0268*** age (0.0121) (0.0120) (0.0121) (0.0103) capital 1.9116*** 1.9045*** 1.9101*** 1.6941*** (0.0749) (0.0730) (0.0750) (0.0614) inventory 0.3612*** 0.3781*** 0.3622*** 0.2730*** (0.0736) (0.0626) (0.0737) (0.0594) 2,600 2,604 2,600 2,600 Obs. Pseudo R2 0.0772 0.0772 0.0772 0.3941 LR Chi2/F-stat 1183.33 1185.88 1183.47 98.79 (1)

ω z age capital inventory Obs. Pseudo R2 LR Chi2/F-stat

Leather Shoes 3.3577* 2.9263 5.4409*** (2.0382) (2.0558) (1.2963) 4.2853* 4.2256* 2.1659 (2.2607) (2.2800) (1.4527) -0.1068*** -0.1002*** -0.1039*** -0.0627*** (0.0172) (0.0170) (0.0173) (0.0112) 2.9326*** 3.0826*** 2.9501*** 1.9356*** (0.1472) (0.1333) (0.1477) (0.0840) 0.2903*** 0.2139*** 0.2654*** 0.1981*** (0.0716) (0.0706) (0.0730) (0.0445) 2,208 2,217 2,208 2,208 0.0674 0.067 0.0677 0.3377 789.57 788.46 793.03 65.7

26