key: cord-0893783-22i7zbil
authors: Trinh, Long Quang; Morgan, Peter J.; Sonobe, Tetsushi
title: Investment behavior of MSMEs during the downturn periods: Empirical evidence from Vietnam
date: 2020-09-25
journal: nan
DOI: 10.1016/j.ememar.2020.100739
sha: 070c9058d95e6143c2017c178edbec8924f7ffc2
doc_id: 893783
cord_uid: 22i7zbil

This paper analyzes panel data of micro, small, and medium-sized enterprises (MSMEs) in Vietnam, covering five two-year periods from 2003 to 2012, in order to understand how MSMEs' investments in their own productive fixed assets and in other firms' equities and real estate increased or decreased after the global financial crisis of 2007–08. It finds that MSMEs increased investments in other firms and real estates and reduced investments in their own businesses considerably after the crisis. This, along with the study's other findings, suggests that the global financial crisis marked the beginning of resource reallocation from hitherto proliferating MSMEs to rapid growth of larger firms in the Vietnamese economy.

In emerging market economies (EMEs), non-financial firms can invest in other firms as well as their own production capacities and/or innovative activities unlike their counterparts in lowincome developing countries that are without stock markets (Demir 2009a (Demir , 2009b . As investors, such firms can contribute to economic growth and shape future industrial structure.

Their contribution could be more important than that of their counterparts in advanced economies, where the financial market is dominated by institutional investors. However, not all emerging-market firms are able and willing to invest, especially during periods of economic downturn when firms tend to face financial constraint. Using firm-level data from EMEs in Central and Eastern Europe, Kolasa et al. (2010) and Burger et al. (2017) revealed that domestic firms, younger firms, smaller firms, export-oriented firms, and service-sector firms tended to cut investment more drastically than foreign-owned, older, and larger manufacturing firms during the global financial crisis (GFC) of 2007/2008, which was said to be the worst financial crisis since the Great Depression of the 1930s until it was dwarfed by the crisis. This finding raises a question as to whether it applies to any other EMEs in East Asia, if not in Europe.

Emerging economies in East Asia sailed over the GFC relatively smoothly because they had learned their lesson from the Asian financial crisis of 1998 and built up financial firewalls (Pempel 2015) . Firms in this region could expect a relatively quick recovery from the economic downturn and high capital gains from investing in financial and real estate assets of which prices had precipitated from a height in the early phase of the GFC.

Among other findings, this study found that MSMEs in Vietnam increased financial investment, including investment in real estate assets, and decreased investment in equipment and other productive fixed assets for their own use during the GFC. Thus, increasing financial investment during periods of economic downturn is not unique to state-controlled firms that were favorably treated by the Chinese Government, as described by Bo et al. (2014) ; rather, it can be classified as common investment behavior among MSMEs in some EMEs. This study also found that, compared with other MSMEs, financial investment is greater among those MSMEs with larger amounts of external finance, experience of engaging in foreign trade, and 12 to 35 employees. This, however, does not mean these firms increased financial investment more than other MSMEs; rather, the increase in financial investment during the economic downturn is not significantly related to any particular firm characteristics.

By contrast, the magnitude of decrease in productive investment due to the economic downturn was found to be significantly and negatively correlated with employment size and the amount of external finance. Specifically, firms with larger employment sizes and external finance reduced productive investment more than other firms, but firm age and other characteristics were not significantly related with the magnitude of decrease in productive investment. These results are consistent with and add new elements to those results concerning investment behavior during the crisis that were obtained by Kolasa et al. (2010) from Polish firm data and differ slightly from the results obtained by Burger et al. (2017, p. 586) from Central and Eastern European firm data. Apart from changes in investment from normal times to the period of downturn, the present paper highlights that the longer-term relationship between productive investment and various firm characteristics over the entire data period is consistent with the common finding in the literature that younger, larger, formal, and exporting firms with more favorable access to external finance invest more.

The remainder of the paper is organized as follows. Section 2 provides a brief literature review; section 3 documents some characteristics of the Vietnamese economy and its MSME sector; section 4 discusses empirical strategy and reports the results of the descriptive analysis of the data; section 5 reports the empirical results; and, finally, section 6 concludes the paper with a summary of the findings and policy implications.

This paper is closely related to two strands of the economic literature on firm investment behaviors. These two strands correspond with two types of firm investment: 1) investment in their production capacity and 2) investment in financial assets. Although firms' decisions on the two types of investment may be interlinked, for the sake of simplicity, this review treats the two strands of the literature separately.

A firm's investment in their production capacity depends on their expected future net cash flow and associated risks, as well as financial constraints. Given this, firms are highly heterogenous. Thus, an immense body of literature has been developed to understand such heterogeneity by linking investment behaviors with firm characteristics, such as firm size, firm age, ownership structure, and external finance (see, e.g., Nugent and Yhee 2002; Diamond 1989; Jensen and Merkling 1976) . The relationship between firm characteristics and financial constraints together with expected growth performance may vary from time to time. Thus, a growing number of studies investigate how firms change their investment behaviors, productivity growth, and entry and exit from normal times to recessions and how such changes are related to firm characteristics. Among important examples are studies by Kolasa et al. (2010) and Burger et al. (2017) , who link changes in firm behaviors with firm characteristics by using firm-level data from Central and Eastern Europe before and during the GFC. The present paper investigates whether a similar linkage between changing investment behaviors and firm characteristics existed in an Asian emerging market before and after the GFC.

Turning to studies of firms' investment in financial assets, it seems useful to begin with Demir's (2009a Demir's ( , 2009b ) studies based on data from Argentina, Mexico, and Turkey. These studies found that de-industrialization took place in the three emerging markets in the 1990s, as firms decreased investment in their own production capacity to increase financial investment when the financial liberalization opened the window of opportunity for alternative investment J o u r n a l P r e -p r o o f Journal Pre-proof in financial assets. It does not follow, however, that the reallocation of firms' investment funds away from their own productivity capacity to financial assets always leads to deindustrialization . This is because, if increased financial investment is mostly equity investments in other firms that have high growth prospects, reallocation can result in creative destruction, as defined by Schumpeter (1934) .

Extending the creative destruction argument, Davis and Haltiwanger (1992) and Caballero and Hammour (1994) , for example, argue and present evidence that economic downturns accelerate the process of "cleansing", where less efficient firms are replaced by more efficient firms. Recent studies, however, point out that recessions exacerbate financial constraints and labor market frictions, making it difficult for more productive firms to replace less productive ones. For example, Hallward-Drimeier and Rijkers (2013) analyzed firm-level data from Indonesia before and after the Asian financial crisis and found that, while new firms that entered during the crisis had relatively high productivity, the crisis induced relatively productive firms to exit. Similarly, Carreira and Teixeira (2016) found that a "non-negligible fraction of high-productivity firms actually shutdown" in Portugal during the GFC. These studies did not find evidence lending strong support to the cleansing hypothesis.

As Carreira and Teixeira (2016) argue, whether the cleansing effect prevails or is reversed by labor and financial market failures may depend on the nature and extent of economic downturn. According to the International Monetary Fund's World Economic Outlook (WEO), the average annual GDP growth rate in Emerging and Developing Europe went down by 12.9 percentage points, from 7.2 percent in 2007 to -5.7 percent in 2009 , and that in Emerging and Developing Asia went down by only 4.6 percentage points, from 11.2 percent to 7.6 percent.

The GFC may not scar the Asian emerging markets that are exacerbating market failures as much as their European counterparts. To the extent this is the case, MSME owners in Asian emerging markets may find the stock price plunges at the beginning of the economic downturn a favorable opportunity to invest in those listed firms they expected to be more profitable in future than their own businesses. This constitutes the basic hypothesis of the current paper.

To our knowledge, the most closely related study is Bo et al. 's (2014) study investigating the investment behaviors of listed firms in China. One of the main results outlined in this study is that non-government firms did not significantly increase financial investment during the GFC, even though state-controlled, non-financial firms did increase financial investment, expecting high capital gains, which is likely because the latter were treated favorably by the financial sector as well as the Chinese Government. We believe this is not the case in some other emerging markets in Asia, especially in Vietnam. Based on the above review of the two strands of literature, one on firms' investment in their own production capacity and the other in their financial investment, we hypothesize the following. First, Vietnamese MSMEs' investment in their own production capacity is related to their observable characteristics in the same way as existing studies have found. Second, the change in this type of investment during the economic downturn caused by the global financial crisis is related to observable characteristics in the same way as Kolsa et al. (2010) and Burger et al. (2017) found. Third, and by contrast, the increase in these firms' financial investment is not related to any observable characteristics. A part of the reason is that their increased financial investment is motivated by the high expected performance of the listed firms, not themselves. The other reason is that their financial investment would be financed mostly by owners' assets, of which abundance would not be associated with firm age, size, formality, or other characteristics. For example, a new, small, informal firm owned by an affluent business person who also owns many other firms may answer to the survey that the firm's owner increased financial investment considerably soon after stock prices plunged.

Journal Pre-proof One may think the second and third hypotheses are contradictory if MSMEs are financially constrained. Indeed, it is possible to think of a situation in which the sum of financial investment and productive investment must be equal to a given amount of investment fund. In such a situation, if productive investment increases with a particular characteristic, financial investment must decrease, as that characteristic takes a larger value and cannot be independent of all characteristics as hypothesized above. However, the two types of investment may face different extents of financial constraint. If MSMEs in emerging markets and developing countries are poorly managed, as attested by Bloom et al. (2019) for example, both financial institutions and MSME owners may be reluctant to finance increases in the working capital or equipment of MSMEs, but they may be willing to finance financial investment that is expected to have high returns and low risks. Therefore, there is no contradiction among the hypotheses above.

Before closing this section, it is useful to elaborate on the first and second hypotheses above because they simply state that the relationship between observable firm characteristics and investment behavior (or its change) is expected to be the same as existing studies have reported. As to firm size, when the economy turns downward, relatively large firms reduce investment in their own production capacity to adjust to the declining demands for their products, whereas smaller firms do not reduce productive investment because they rarely invest in production capacity even in normal times. 1 With regard to firm age, we follow Das (1995) to hypothesize that the investment behavior of relatively old firms during a downturn period does not differ from that of relatively young firms with otherwise similar characteristics, business performance, and financial conditions. 2 As to external finance, MSMEs with more external finance invest more during normal periods and reduce investment more during downturn periods than otherwise similar MSMEs with less external finance. This is because high debt makes borrowing expensive, especially in downturn periods, as argued by Petersen and Rajan (1995) . 3 As to formality, we hypothesize that formal MSMEs are already so well-equipped that their investment rates tend to be lower in normal periods and even lower in downward periods, compared with informal firms with, for example, similar employment sizes, ages, and amounts of external finance. 4

Since the opening up of the economy in 1988, Vietnam has enjoyed high economic growth in most years. During the 1990s, the growth rate remained around 7% per year and the poverty rate reduced from 58% to 25%. During the period of 2000-08, the economy grew at 7. 5-7.8% per year, except for year 2007 -the year it joined the World Trade Organization (WTO) -and grew at 8.5%. Triggered by the global financial crisis of 2007-08, however, an economic downturn started in 2008. The growth rate fell to 5.8-6% per year. The largest sector in the economy, the industry and construction sector, which accounted for 40% of GDP and was viewed as the engine of the country's growth, suffered the most. The average growth rate of this sector declined from 10.2% in 2000-2006 to 7% in 2007-2011. Several factors are considered to be responsible for the slowdown. From 1999 to 2007, the country adopted an expansionary policy to support high growth by means of rapid investment. This policy created inflationary pressure and aggravated inefficiencies across all sectors of the economy. Just before the global financial crisis began, Vietnam joined WTO and experienced a huge surge of foreign direct investment (FDI) inflows, and the impact on the economy's stability was not sufficiently neutralized due to policymakers' lack of experience.

Joining WTO implies that the negative impact of the global financial crisis was transmitted quickly to the economy. Subsequently, FDI inflows decreased substantially and export growth slowed down considerably, both of which had strong impacts on different sectors of the economy. For export-oriented industries, the major cause of slowdown is said to be the decline in foreign demand and the trade barriers imposed by some trade partner countries. For importsubstituting industries, the major cause is said to be intensified competition in domestic markets, partly due to the removal of import tariffs as a consequence of joining WTO. The economic slowdown after the global financial crisis is said to have been prolonged due to the mismanagement of macroeconomic policies, such as an inconsistency between fiscal and monetary policies and untimely stop-go swings.

As in other developing countries, MSMEs are a major driver of economic development in Vietnam. In 2017, formally registered MSMEs contributed around 32% of GDP and 36-38% of the country's total investment. The General Statistics Office (GSO) of Vietnam reported that 507.9 thousand firms out of a total of 517.9 thousand formal enterprises in the economy are MSMEs (GSO 2018). The average growth rate of this formal MSME sector is approximately 14 percent per year. Turning to informal MSMEs, probably 74% out of a total of 5.1 million "individual establishments" with 8.7 million persons engaged (32.3% of the total employment in the country) are informal while the remaining 26% are formally registered as "household businesses" (GS0 2018). Thus, MSMEs are very important players in this economy.

MSMEs in Vietnam are characterized by limited access to finance, limited business skills and managerial capabilities, low levels of technological advancement, and inadequate support from the government. For example, nearly 80% of MSMEs surveyed in a study (CIEM, DoE, ILSSA, 2014) financed their investment projects by using internal rather than external sources.

The vast majority of them are micro and small enterprises, while medium-sized enterprises tend to have more favorable access to finance. Thus, the panel data cover the five two-year periods from 2003 to 2012. The questionnaire used in the surveys includes questions about firms' general characteristics, production and costs, sale revenues, export activities, employment, investments, assets, liabilities, and taxes in the previous two years.

After each survey round, the sample firms that had ceased operation or stopped cooperating with the survey team were replaced by both formal MSMEs selected from a list compiled by the GSO and informal MSMEs selected in each province. Thus, the number of firms remained at an almost constant 2,500 for every survey round. Not all sample firms answered all the questions, however. Only about 1,500 sample firms per survey round provided the information necessary to carry out the regression analysis specified below.

As mentioned earlier, most empirical studies of firms' investment behaviors use either the Q model or the accelerator model (e.g. Lensink and Sterken 2000; Bo and Zhang 2002) . While the panel data we use do not have sufficient data points in the time series direction to construct Q for each sample firm, the data do allow us to estimate an accelerator model, which may be written as follows:

(1)

where is the either productive or financial investment rate of firm i at period t (t = 2, …, 5), is the logarithm of the number of full time workers; is firm age as of the first survey round; is the logarithm of the amount of external finance; is a dummy variable indicating whether firm i is formally registered or not; is a dummy variable that is equal to 1 if period t was either period 4 or period 5, when the Vietnamese economy was caught in the economic downturn and 0 otherwise; is a vector of accelerator variables as explained below; is a vector of control variables also explained below; is a firm fixed effect; is a time effect common to all firms; and is an error term. Coefficients to are scalers and and are vectors. Equation (1) and an explanatory variable. To mitigate this problem, we include in vector X those variables representing export markets and foreign direct investments as well as industry dummies and province dummies that indicate which industry and province a firm belonged to. 5

As we will see below, however, a number of observations have no productive investments; that is, the dependent variable is censored from below at zero and has frequent zero-valued observations. The regression analysis below applies the random effect (RE) Tobit estimator to cope with this problem as well as the fixed effect (FE) estimator to the panel data to check the robustness of the results of the FE estimation.

We are particularly interested in estimating the coefficients on Size, Age, Fin and Formal, and the coefficients on the terms that interact these variables with the economic downturn dummy because these coefficients are closely related to the hypotheses discussed in Section 3 above.

The inclusion of the square of Size is intended to capture the concavity or convexity of the relationship between the investment rate and employment size. The firm age variable is included as the age at a particular point in time because all firms grow older at the same pace.

The effect of aging from period 2 to period 3, if any, is absorbed by time effect or period dummy. Since the cross-section variation of firm age as of time 1 is absorbed by the firm fixed effect , coefficient cannot be estimated with the fixed effect model specification (i.e., within estimator). Still it could be estimated with the random-effect model specification. 5 We categorize firms into seven industries: food-processing, garment and textiles, wood and furniture, chemicals, non-metal materials, machinery and other industries based on the 2-digit code in the Vietnam Industrial Code 1993, which corresponds with ISIC 3 revision.

The accelerator investment model could be specified in different ways in empirical analyses. We adopt a standard specification highlighting sales revenue and profit, following the lead of Harris et al. (1994) , Lensink and Sterken (2000) , Lensink et al. (2001) , and Bo and Zhang (2002) among others. As Harris et al. (1994) argue, sales revenue is one of the major determinants of a firm's investment decisions because it is a signal of the size of market demand for the firm's products and services. As Lensink et al. (2001) argue, retained profit should also play an important role in investment decisions because it is a signal of future profitability and is related to the availability of internal funding for investment. Thus, vector in our model consists of the ratio of sales revenue in period t to fixed asset , and the ratio of retained profit to fixed asset

We now turn to the specification of regression analysis regarding the portfolio choices made by MSMEs between productive and non-productive investments. To capture this choice, we consider the proportion of productive investment to the sum of productive and non-productive investments in period t. This proportion may be explained by the same variables that appear on the right-hand side of equation (1), that is, lagged values of employment size, age, external finance, and formality, downturn dummy , interaction terms, , firm fixed effects, and period dummies, which may be denoted by vector . In other words, we consider a regression with the percentage of productive investment as the dependent variable and as the explanatory variables.

The estimation of this equation, however, may suffer from a serious selection bias. This is because a number of the sample MSMEs did not invest at all, which makes the denominator of the dependent variable zero, thereby making it impossible to include these enterprises in the analysis.

To the extent that the portfolio choice between productive and non-productive investments is related to the choice of undertaking non-zero investment, ignoring those enterprises with zeroinvestment leads to a selection bias. To address this concern, we include the inverse Mills ratio in the equation explaining the percentage of productive investment. The Mills ratio can be obtained by J o u r n a l P r e -p r o o f estimating the probit model that predicts the probability of positive investment. This probit model may be written as follows:

where is the cumulative distribution function of the standard normal distribution, and is the vector of coefficients to be estimated. We estimate the probit model period-by-period following the suggestion of Wooldridge (1995) . This is why coefficients in equation (2) have subscript t,

allowing that the coefficients vary over time. The estimated coefficients are used to calculate the inverse Mills ratio, The next step is to estimate the fixed-or random-effect model of the percentage of productive investment, denoted PPI:

, (3) where is an error term. Equation (3) has the firm fixed effect and the time effects as well.

They, however, are not explicitly expressed in the equation and instead included in . [2003] [2004] ; period 2 is years [2005] [2006] ; period 3 is years [2007] [2008] ; period 4 is years [2009] [2010] ; and period 5 is years [2011] [2012] It is interesting to observe that the percentage of sample firms with positive investments including non-productive ones was higher in periods 3, 4, and 5 (52.9% and 44.7%), when the economy was in a downturn, than in period 2 before the global financial crisis (39.4%). By contrast, the percentage of sample firms with positive productive investments as well as the productive investment rate was considerably lower in periods 4 and 5 than in periods 2 and 3. Consistently, the percentage of productive investment to total investment declined from 85.7% in period 2 to 35.1% and 41.2% in periods 4 and 5.

This proportion reduced drastically as early as in period 3. Presumably the reason is explained partly by the booming stock market in Vietnam, where the average stock market index grew from 390 in period 2 to 745 in period 3 and then went down to 498 and 425 in periods 4 and 5, respectively. That is, MSMEs invested in other firms' equities, benefitting from a surge in prices.

Another factor may be the surge of real estate prices.

The average size of firm in terms of the number of employees is about 14. Although not shown in the table, 75% of the sample enterprises are micro enterprises -enterprises with less than 10 employees. The average years of operation is about 20 years, with 56.54% of firms older than 12 years old. More than 30% of the firms in the sample had access to external finance. The percentage of firms with external finance was very low in period 2 but almost constant in other periods. The average amount of external finance did not vary much from period to period.

The average values of the accelerator variables -that is, sales revenues/fixed assets and profits/fixed assets -decreased only a little in the crisis period and the subsequent downturn periods. This finding lends support to the view that Asian emerging economies were rather insulated from the global financial crisis, even though FDI inflows and export growth decreased considerably.

Presumably, the reason why sales revenues and the profits of firms in the sample did not decrease much was that less than 8% of sample firms were engaged in external trade and FDI inflows decreased in period 3, but returned to the previous level quickly in periods 4 and 5. Table 2 presents the fixed-effect model estimates of equation (1). The coefficients on the three period dummies in column 1 indicate that the productive investment rate went down in period 3, when the global financial crisis took place, and that it went down further in the subsequent downturn periods. 6 Notes: Dependent variable is productive investment rate. Standard errors in brackets; * p < 0.1, ** p < 0.05, *** p < 0.01. Period 3 is a dummy variable which is equal to 1 if period is years [2007] [2008] and 0. Similarly, Period 4 indicates whether the period is years [2009] [2010] , and Period 5 indicates whether the period is years 2011 -2012 . The first period (2003 data are used for lagged variables. The second period (years [2005] [2006] ) is used as the reference point. D is a dummy variable equal to 1 if the period is either the fourth or the fifth period, and equal to 0 otherwise. Firm age is time-invariant and equal to the age when firms participated in the first survey. The amount of external finance and FDI inflows are real values in the 2005 constant Vietnamese Dong. In all specifications, the industry and province that a firm belongs to are controlled by industry dummies and location dummies. Source: The University of Copenhagen-CIEM-ILSSA data and the authors' calculation Columns (2) to (5) include one of the four interaction terms of interest whereas column (6) includes all of these interaction terms. Column (6) shows our benchmark estimation result.

The interaction of the downturn dummy D and employment size in columns (2) and (6) has a negative and highly significant coefficient, but the coefficient on employment size in these J o u r n a l P r e -p r o o f columns is insignificant. These results are consistent with the hypothesis that larger enterprises tend to decrease their investment rates more than smaller ones in a downturn period. The coefficient on the interaction of D and firm age in columns (3) and (6) is not significantly different from 0, which suggests that firms, regardless of firm age, tend to be reluctant to invest in productive assets when the economy turns downwards.

While the interaction of D and external finance has a negative and highly significant coefficient in both columns (4) and (6), the coefficients on external finance in these columns are positive and highly significant. These results lend strong support to the hypothesis that firms with higher debt tend to invest more in normal periods but reduce productive investment in a downturn period. The coefficient on the interaction of D and the formal firm dummy in columns (5) and (6) is negative and highly significant, and that on the formal firm dummy in these columns is also negative and highly significant. These results are consistent with the hypothesis that formal firms are so well-equipped already that their investment rates are lower in normal periods and even lower in downward periods, compared with informal firms that are otherwise similar.

The estimated coefficients are stable across columns (1) to (6). One of the two accelerator variables has a highly significant coefficient, which suggests that the choice of the accelerator model is not invalid. One might wonder why the coefficient on the foreign trade dummy is consistently negative and significant. Although we are not sure about the reason, our conjecture is that MSMEs engaged in export markets specialize in less capital-intensive areas of trade in accordance with their additional income from foreign markets and hence invest less than MSMEs catering for domestic markets.

(1) (2) (3) (4) (5) Table 3 reports the result of random-effect Tobit estimation of equation (1). This estimation takes the mass of zero-investment observations into account, but it treats the firm fixed effect as a random variable and is inefficient in the statistical sense. As in Table 2 , the coefficients on the period dummies in column (1) of this table indicate that the decline in the investment rate began in period 3, when the stock market in Vietnam was booming, because financial resources were diverted away to financial assets and real estates. The estimates shown in Table 3 are generally similar to those in Table 2 . This is true especially for the estimated coefficients on the interaction terms. Thus, the results are again consistent with our hypotheses.

There are some qualitative differences between the results reported in Tables 2 and 3, however. The coefficients on employment size and its squared term are statistically insignificant J o u r n a l P r e -p r o o f in Table 2 but significant in Table 3 . The positive coefficient on employment size and the negative coefficient on the size squared in Table 3 suggest that the investment rate increases with size until the number of employees reaches six persons and then decreases as the size increases further. Similarly, the coefficients on the urban dummy and FDI inflows are insignificant in Table 2 but negative and highly significant in Table 3 . These increases in significance levels are considered to result either from correlations between these explanatory variables and the unobserved firm fixed-effect or from the better treatment of the zeroobservations problem or from both. In any case, these significant estimates suggest that both very small and relatively large MSMEs, especially those located in urban areas and/or provinces receiving large FDI inflows, tend to have smaller investments in productive assets.

Tables 4 and 5 present the fixed-effect estimates and the random-effect Tobit estimates of equation (1), where financial investment rate serves as the dependent variable. These tables are structured similar to tables 2 and 3. Column 6 in both tables highlights our preferred results.

The results show that the coefficients on the three period dummies in Column 1 indicate the financial investment rate increased in periods 3 and 4 and then declined in period 5 when the economy began to recover from the period of downturn. This result was found regardless of the estimation methods used. This result further strengthens our previous conjecture that financial resources of many manufacturing firms were diverted to financial assets and real estate. There exist some differences between the results reported in tables 4 and 5. While the results obtained from the fixed effect estimator suggest firms tended to increase their financial investment rate during the downturn and subsequent periods regardless of firm size, firm age, external finance and formality of firms, the random-effect estimates indicate that larger firms and older firms were likely to reduce their financial investment rate during the downturn period. These results also indicate that firms with higher debt are more likely to invest in financial assets during the downturn period. The coefficients on employment size and its squared term are not statistically significant in Table 4 but significant in Table 5 . The positive coefficient on employment size and the negative coefficient on the size squared in Table 5 suggest that, similar to productive investment rate, financial investment rate also increases with firm size and then begins to decline when firm size reaches approximately six or seven people.

Similarly, the coefficients on firm age, and the formal firm dummy are not statistically significant in Table 4 but highly statistically significant in Table 5 . These increases in significance levels are considered to result either from correlations between these explanatory variables and the unobserved firm fixed-effect or from the better treatment of the zeroobservations problem, or from both. In any case, compared with other MSMEs, these significant estimates suggest financial investment is greater among those MSMEs with larger amounts of external finance, experience of engaging in foreign trade, and 12 to 35 employees. 

The results of the probit model estimation of equation (2) are presented in Table 6 . The major purpose of estimating this equation is to obtain the inverted Mills ratio (IMR), which is used to correct possible sample selection bias in estimation of the function that explains the proportion of productive investment to total investment, which we will turn to in the next sub-section.

According to Wooldridge (1995) , the sample selection bias in panel data analysis is mitigated more effectively if IMR is obtained by estimating the probit model in a period-by-period manner.

Thus, the vector of coefficients in equation (2), , has subscript t, and Table 6 has four columns, each reporting the estimated probit model within one period. The estimated IMR is given by the following formula:

where ( ̂ ) is the standard normal density, ( ̂ ) is the standard cumulative distribution, and ̂ is the vector of estimates reported in Table 6 . 2017 Notes: Standard errors in brackets; * p < 0.1, ** p < 0.05, *** p < 0.01. In all specifications, industry dummies, and location dummies are included. Period 3 is a dummy variable which is equal to 1 if period is years 2007-2008 and 0. Similarly, Period 4 indicates whether the period is years [2009] [2010] , and Period 5 indicates whether the period is years 2011 -2012 . The first period (2003 data are used for lagged variables. The second period (years [2005] [2006] is used as the reference point. The amount of external finance and FDI inflows are real values in the 2005 constant Vietnamese Dong. In all specifications, the industry and province that a firm belongs to are controlled by industry dummies and location dummies. Source: The University of Copenhagen-CIEM-ILSSA data and the authors' calculation Some of the estimates reported in Table 6 are noteworthy. First, just as in Tables 3 and 5, the employment size and its squared term have positive and negative coefficients, respectively, and the coefficients on both the urban dummy and FDI inflows are negative. Throughout the four periods, these coefficients are highly significant. Thus, both very micro-sized firms and SMEs in urban areas or provinces competing with foreign firms tend to have relatively low propensity to invest. Second, just like Tables 2 and 3, Table 6 reports that external finance is positively associated with the likelihood of positive investment, as one would expect.

Tables 7 and 8 shows the fixed-effect estimates and the random-effect Tobit estimates of equation (3), respectively. These tables are designed to look similar to Tables 2 and 3 with the only difference being the inclusion of the inverse Mills ratio. Again column (6) is our favorite result. Note, however, the dependent variable in Tables 5 and 6 is percentage, not fraction, whereas that in Tables 2 and 3 is usually less than 1 but can exceed 1. Thus, the magnitude of coefficients in these two sets of tables cannot directly be compared. Notes: Dependent variable is the proportion of productive investment expressed in percentage. Standard errors in brackets; * p < 0.1, ** p < 0.05, *** p < 0.01. Period 3 is a dummy variable which is equal to 1 if period is years 2007-2008 and 0. Similarly, Period 4 indicates whether the period is years [2009] [2010] , and Period 5 indicates whether the period is years 2011 -2012 . The first period (2003 data are used for lagged variables. The second period (years [2005] [2006] is used as the reference point. D is a dummy variable equal to 1 if the period is either the fourth or the fifth period, and equal to 0 otherwise. Firm age is time-invariant and equal to the age when firms participated in the first survey. The amount of external finance and FDI inflows are real values in the 2005 constant Vietnamese Dong. Inverse Mills ratio is obtained by using the estimated probit model shown in Table 4 . In all specifications, the industry and province that a firm belongs to are controlled by industry dummies and location dummies. Tables 2 and 3, this is not so in Tables 5 and 6 . These results are obtained because there was a massive increase in financial and speculative investments in the third period, which was followed by the sizable decline in productive investments in the ensuing downturn periods.

Unlike Tables 2 and 3, Tables 7 and 8 report insignificant coefficients on the interaction terms throughout the columns. These results, together with results presented in Table 4 and 5, indicate that the proportions of productive asset investment were considerably lower after the global financial crisis than before for most MSMEs across the board -small or large, young or old, formal or informal, debt-laden or unable to borrow.

While the coefficient on external finance in Tables 2, 3 and is positive and significant, that in Tables 7 and 8 is negative and significant. These results lend support to the view that MSMEs used external finance more for the purpose of non-productive investments than for productive investments over the entire periods of study. As to the urban dummy, FDI inflows, and the foreign trade dummy, the qualitative results of Table 5 are similar to those of Table 2 , and those of Table 6 resemble those of Table 3 . These results suggest that the participation of MSMEs in foreign trade is negatively associated with productive investment, which may be related with specialization in accordance with comparative advantage, and that MSMEs located in urban areas or provinces with large FDI inflows tend to undertake less productive investment or more non-productive investment, probably because MSMEs surviving in congested urban areas and those catering for foreign firm subsidiaries are less capital-intensive.

Finally we would like to add that we regard Tables 2 and 5 as the main empirical results of   this paper and Tables 3 and 6 are primarily intended to check their robustness to a change in estimator, even though some results in the latter two tables are informative, as discussed above.

Further robustness checks, including robustness of the main results to changes in sample and to the replacement of some explanatory variables by related but different variables, are provided in the Appendix (Appendix Tables A1 to A4).

Our analysis of the panel data of about 2,500 Vietnamese MSMEs covering the period [2003] [2004] [2005] [2006] [2007] [2008] [2009] [2010] [2011] [2012] has revealed that their investments in productive assets relative to the stock of such assets declined drastically after the global financial crisis of , and that this reduction in rates of productive investment was larger for those firms that had relatively large employment sizes and J o u r n a l P r e -p r o o f those formally registered as companies than for relatively small and informal ones. It is also found that MSMEs' non-productive investments in other firms' equities and real estates increased substantially in 2007 and early 2008 probably because of the then booming stock market even though such investment decreased back to the previous level during the ensuing period of economic downturn. According to our regression results, external funds were used to finance both MSMEs' productive and non-productive investments but contributed to increasing non-productive investments more than productive ones.

The resulting analysis of the data suggests that MSMEs found neither the global financial crisis nor the economic downturn favorable for productive investments, such as in greater mechanization or general business expansion, and turned instead to financial or speculative investments in more dynamically growing listed companies and real estates. This does not mean that all MSMEs, without exception, behaved in this way. On the contrary, a number of innovative MSMEs in Vietnam were able to take advantage of economic downturn. Our data analysis captures the overall tendencies. It is noteworthy that not only innovative firms but also many other MSMEs that increased non-productive investments can be winners and contribute to economic growth because the latter find it more profitable to reallocate resources away from their relatively inefficient use to more rapidly growing firms.

This trend for resource reallocation, or structural transformation, was activated by finance and trade liberalization and especially driven by entry in the WTO, which exposed many MSMEs to greater competition from imports, thereby discouraging them from expanding their own businesses. Further liberalization will deliver benefits to the country, at least in the long run. A caveat is that it may overheat speculative investments and lead to an asset-inflated bubble economy along with rampant market monopolization. Thus, further liberalization in finance and trade ought to be accompanied by prudent macroeconomic management and effective enforcement of antitrust and competition policy.

J o u r n a l P r e -p r o o f Several attempts to check the robustness of our major findings are made as shown in Appendix   Tables A1 to A4. Appendix Table A1 is intended to check the robustness of the results shown in Table 2 by using a balanced panel. That is, the fixed-effect model estimation of equation (1) is carried out by using only the sample firms without any missing data for any variables specified for any period. As a result of eliminating those firms without complete data, the number of observations reduced from 7733 to 3760. The estimation results remain qualitatively similar to the results presented in Table 2 .

Appendix Table A2 reports the robustness check of the results shown in Table 3 by using a balanced sample. Similar to the Appendix A1, the estimation results presented in Appendix Table 2 are also qualitatively consistent with the results presented in Table 3 . This suggests that although there is some quantitively difference between balanced and unbalanced samples, our estimates have generally not suffered from attrition bias.

Appendix Tables A3 and A4 are used to check the robustness of results presented in Tables 2 and 5, by using the different but related variables. More specifically, we replace the log of employment size with a dummy variable ("SME dummy") which is equal to 1 if the firm has more than ten workers and hence is not micro-sized. Firm age is also replaced by older firm dummy which is equal to 1 if firm age is 15 years or above and 0 otherwise (15 years is the median age in our sample). The log of the amount of external finance is replaced by external finance dummy which is equal to 1 if the firm has external finance and 0 otherwise.

Appendix Table A3 shows that SME firms (relatively large MSMEs) tend to decrease their investment rates more than micro firms in downturn periods. Meanwhile, there is no difference in productive investment rate between younger firms and older firms in downturn periods.

Firms with external finance tend to invest more in a normal period but reduce productive investment in a downturn period. Finally, formal firms have lower investment rates in both normal periods and (even lower) in downward periods. The results in Appendix Table A4 J o u r n a l P r e -p r o o f suggest that changes in firms' portfolio choice occurred in the same manner regardless of their characteristics. This suggests that the empirical results presented in Table 2 and Table 5 are reliable regardless of measurement of each firm's characteristics. 

The determinants of financing obstacles

Small and medium enterprise dynamics in Indonesia

Growing firms and the credit constraint

What drives differences in management practices

The impact of uncertainty on firm investment: Evidence from machinery industry in Liaong province of China

Corporate investment during the financial crisis: Evidence from China

Determinants of firm performance and growth during economic recession: The case of Central and Eastern European countries

The Cleansing Effect of Recessions

Entry and exit in severe recessions: lessons from the 2008-2013 Portuguese economic crisis

Accessing Vietnam's socio-economic development after 5 years joining WTO, Finance Publishing House: Hanoi Central Institute for Economic Management (CIEM, Department of Economics at the University of Copenhagen (DoE,) and Institute for Labor Studies and Social Affairs (ILSSA)

What really happens to small and medium-sized enterprises in a global economic recession? UK evidence on sales and job dynamics

The dynamics of employment growth: New Evidence from 18 Countries

Size, age and firm growth in an infant industry: The computer hardware industry in India

Gross job creation, gross job destruction, and employment reallocation

Financial liberalization, private investment and portfolio choice: financialization of real sector in emerging markets

Capital market imperfection and financialization of real sector in emerging markets: private investment and cash flow relationship revisited

Reputation acquisition in debt markets

Did the East Asian crisis disproportionately hit small businesses in Korea

Informal firms and financial inclusion: status and determinants

Financial flexibility across the euro area and the UK. ECB Working Paper, No. 1630

How firms respond to business cycles: The role of firm age and firm size

Firms as liquidity providers: Evidence from the 2007-2008 financial crisis

Kết quả chính thức Tổng điều tra Kinh tế năm 2017

Monetary policy, business cycles, and the behavior of small manufacturing firms

Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia

Legal form, growth and exit of West German firmsempirical results for manufacturing, construction, trade and service industries

The effect of financial liberalization on the capital structure and investment decisions of Indonesian manufacturing establishments

The economic crisis and its effects on SMEs

Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure

Firms in the great global recession: The role of foreign ownership and financial dependence

Why did some firms perform better in the global financial crisis

Capital market imperfections, uncertainty and corporate investment in the Czech Republic

Investment, capital market imperfections, and uncertainty: Theory and empirical results

Trade credit, bank credit and financial crisis

Winners and Losers during a Deep Economic Crisis: Firm-level Evidence from Indonesian Manufacturing

Small and medium enterprises in Korea: Achievements, constraints and policy issues

Micro and small enterprise growth in Turkey: Under the shadow of financial crisis

The effect of credit market competition on lending relationships

Who becomes an entrepreneur? Labor market prospects and occupational choice

Firm entry and employment dynamics during the great recession

SME export performance in Indonesia after the crisis

Employment and SMEs during crises

Is entrepreneurship necessarily good? Microeconomic evidence from developed and developing countries

Selection corrections for panel data models under conditional mean independence assumptions

The role of trade credit in the recent subprime financial crisis

Notes: The same notes as those of Table 2 apply. The sample used here consists of only those firms with no missing data of the variables in the model for each period

The sample used here consists of only those firms with no missing data of the variables in the model for each period

The log of employment size is replaced by SME dummy which is equal to 1 if the firm has more than ten workers and hence is not micro-sized. Firm age is replaced by older firm dummy which is equal to 1 if firm age is 15 years or above and 0 otherwise. The log of the amount of external finance is replaced by external finance dummy which is equal to 1 if the firm has external finance and 0 otherwise. Source: The University of Copenhagen-CIEM-ILSSA data and the authors' calculation

The same notes as those of Appendix Table 3 apply. Inverse Mills ratio is obtained by using the estimated probit model shown in Table 4. Source: The University of Copenhagen-CIEM-ILSSA data and the authors' calculation

We would like to thank editors and anonymous reviewers, Ardeshir Sepheri, Trang T. Le and participants at Chu Hai conference on International Business and Economic