Corporate Cash Holding in Asia

In this paper, we analyze the determinants of corporate saving in the form of changes in the stock of cash for 11 Asian economies using firm-level data from the Oriana Database for the 2002-2011 period. We find some evidence that cash flow has a positive impact on the change in the stock of cash, which suggests that Asian firms are borrowing constrained and that they save more when their cash flow increases so that they will be able to finance future investments. Moreover, we find in the developed economy sample that, as expected, cash flow has a positive impact on the change in the stock of cash only in the case of the smallest firms, which are more likely to be borrowing constrained, and find in the developing economy sample that, as expected, the positive impact of cash flow on the change in the stock of cash declines with firm size. In addition, we find that the cash flow sensitivity of cash declined after the global financial crisis. Finally, we find some evidence that Tobin's q has a positive impact on the change in the stock of cash.


INTRODUCTION
Corporate saving has always played an important role, but it has shown a substantial upward trend in many, if not most, of the developed and Asian economies for the past 2 decades, as pointed out by Karabarbounis and Neiman (2012), and thus is of even more importance now. Yet most previous analyses of saving have focused on household (or personal) saving or national (or domestic) saving, and there have been relatively few analyses of corporate (or firm) saving (refer to the papers in the references section for a representative sampling of previous theoretical and empirical analyses of corporate saving).
Why do firms accumulate liquid assets? In other words, why do firms channel their cash flow into liquid assets rather than into physical capital (capital formation) or into shareholder distributions (dividends)? The high saving rates in Asia have been called a "saving glut" and have been blamed for the pre-global financial crisis "global imbalances," and both academics and policy makers have expended much effort in trying to understand the saving behavior of this region generally (see, for example, Horioka and Terada-Hagiwara (2011), but the factors driving the surge in corporate saving in Asia have yet to be disentangled. The objective of this paper is to fill this void by doing an empirical analysis of the saving behavior of Asian firms.
The recent literature on corporate saving has proposed two main explanations for the excess corporate saving in emerging markets. First, emerging markets have a limited supply of financial assets and are financially constrained (see, for example, Dooley, Folkerts-Landau, and Garber 2005;Matsuyama 2007;Ju andWei 2006, 2010;and Caballero, Farhi, and Gourinchas 2008). In this explanation, the underdeveloped financial sector serves as the main driver of corporate saving behavior.
The second strand of the literature on corporate saving focuses on the role of the precautionary motive. In this explanation, excess saving and thence net capital outflows result from precautionary saving arising from idiosyncratic risk (see, for example, Mendoza, Quadrini, and Rios-Rull 2009, Sandri, 2010.
In these precautionary saving models, rising uncertainties cause a decline in corporate investment, as has been particularly noticeable in some of the Southeast Asian economies.
The goal of this paper is not to directly address the macroeconomic phenomenon of the saving glut, but rather to understand the process whereby firms accumulate liquid assets or saving by analyzing firm-level income statement data. In particular, we are interested in why firms channel their cash flow into liquid assets (defined to include cash as well as other financial assets) rather than into physical capital. We also seek to determine whether and when corporate saving behavior can serve as a useful indicator of the extent to which firms face external borrowing constraints. Thus, we focus on two specific determinants of corporate saving: income uncertainty and the cost of external finance.
In this paper, we analyze the determinants of corporate saving in the form of changes in the stock of cash for 11 Asian economies using firm-level data from the Oriana Database for the 2002-2011 period. Figure 1 shows trends in our measure of corporate saving during the sample period. The change in the stock of cash (as a share of total assets) was positive throughout the sample period, which indicates that cash holdings have been increasing throughout this period. This supports the view that there was a "saving glut" in Asia. However, some variations can be observed across time and among groupings of economies. Firms in our sample continued to save more cash until 2010 despite the slight dip in 2008, and the cash saving rate peaked in 2010 before dropping quite sharply in 2011. This general trend can be observed in both developed and developing economies, but firms' cash holdings increased significantly more in developing economies than in developed economies after 2007, until they dropped sharply in 2011.
To preview our main findings, we find some evidence that cash flow has a positive impact on the change in the stock of cash, which suggests that Asian firms are borrowing constrained and that they save more when their cash flow increases so that they will be able to finance future investments. Moreover, we find in the developed economy sample that, as expected, cash flow has a positive impact on the change in the stock of cash only in the case of the smallest firms, which are more likely to be borrowing constrained, and find in the developing economy sample that, as expected, the positive impact of cash flow on the change in the stock of cash declines with firm size. In addition, we find that the cash flow sensitivity of cash declined after the global financial crisis. Finally, we find some evidence that Tobin's q has a positive impact on the change in the stock of cash.
The remainder of the paper is organized as follows. We discuss theoretical considerations in Section II, the estimation model and the estimation method in Section III, the data source in Section IV, the estimation results in Section V, and the summary and conclusions in Section VI.

II. THEORETICAL CONSIDERATIONS
In this section, we consider the determinants of corporate saving from a theoretical perspective. Our discussion draws heavily on the analyses on Almeida, Campello, and Weisbach (2004), Khurana, Martin, and Pereira (2006), Riddick and Whited (2009), and Almeida, et al. (2013) (see also Karabarbounis and Neiman (2012) and Huang (2011), who also discuss the importance of borrowing constraints as a motive for corporate saving).
Households and firms are very different economic entities with very different objective functions, with households consuming in order to maximize their utility and firms investing in plant and equipment and using that plant and equipment in order to produce goods and services and make profits. However, there are many similarities between the two types of economic entities nonetheless. For example, in the same way that households save in order to finance their future consumption, firms save in order to finance their future investment. Moreover, in the same way that households have a choice between financing their consumption by borrowing or by drawing down their previously accumulated saving, firms have a choice between financing investment by borrowing or by drawing down their previously accumulated saving. Furthermore, in the same way that borrowing constraints and borrowing costs will influence households' decisions about how to finance their consumption, borrowing constraints and borrowing costs will also influence firms' decisions about how to finance their investment.
As the Fisher Separation Theorem states, the firm's decision regarding physical investment is, of course, paramount, and once this decision is made, the firm will then decide how to finance the investment, whether it be from internal funds (cash) and/or external funds (loans and bond and equity issues). It is in this context that firms will decide whether and how much financial assets (cash) to hold, and it is this decision upon which we will focus in this paper.
If firms did not face borrowing constraints, they could finance their investment in plant and equipment by borrowing as much as they needed from external sources at market rates and therefore would not need to hold financial assets in preparation for the sudden and unexpected appearance of profitable investment projects (projects with a positive net present value). However, if firms faced borrowing constraints, as a result of which they had to pay more than the market rate of interest when borrowing from external sources, they might choose to hold at least some financial assets at all times to ensure that they were able to take advantage of any profitable investment projects that might suddenly and unexpectedly arise without having to borrow from external sources, thereby saving on expected future financing costs.
Note, however, that holding financial assets confers costs as well as benefits.
One cost of holding financial assets is that doing so reduces the amount of current investments that can be financed from internal sources, and in the case of borrowing constrained firms, this in turn may reduce the amount of current investments the firm can do. Another cost of holding financial assets is that the firm must pay taxes on the interest income earned on holdings of financial assets. Thus, firms must balance the costs of holding financial assets (the cost of foregone current investment projects and the tax liability on the interest income accruing to holdings of financial assets) against the benefits of doing so (the reduction in expected future financing costs) when deciding how much of their assets to hold in the form of financial assets.
We turn next to what these theoretical considerations imply concerning the "cash flow sensitivity of cash," the impact of cash flow on how much firms save in the form of cash, where cash is construed broadly to include all liquid financial assets.
According to the analysis of Almeida, Campello, and Weisbach (2004), an increase in cash flow will cause firms to want to increase current investment as well as future investment. A firm that is not borrowing constrained will not necessarily increase its cash holdings in response to an increase in its cash flow and may use the entire increase in its cash flow to finance current investment because it knows that it will be able to finance future investment using external funds without any difficulty. Thus, an increase in cash flow will not have a systematic impact on how much unconstrained firms save in the form of cash (that is, the so-called "cash flow sensitivity of cash" of unconstrained firms would be expected to be zero).
However, a firm that is borrowing constrained will use at least part of the increase in its cash flow to increase its cash holdings so that it will be able to increase its future investment without relying on external funds. Thus, an increase in cash flow will cause borrowing constrained firms to save more in the form of cash (that is, the "cash flow sensitivity of cash" of borrowing constrained firms will be positive).
By contrast, Riddick and Whited (2009) extend the theoretical analysis of Almeida, Campello, and Weisbach (2004) and show that the cash flow sensitivity of cash may be negative in the case of borrowing constrained firms. They obtain a very different prediction regarding the sign of the cash flow sensitivity of cash because they assume that the firm faces positively serially correlated productivity shocks. As a result, a positive productivity shock will cause the firm's cash flow to increase and its capital to become more productive, and its productivity will revert to its mean only slowly. This will induce the firm to shift its assets from cash to physical capital, and if this substitution effect is strong enough to offset the income effect identified by Almeida, Campello, and Weisbach (2004) and discussed above, the firm will invest and draw down its cash holdings in response to an increase in cash flow caused by a positive productivity shock.
Thus, an increase in cash flow will cause firms to save less in the form of cash (that is, the "cash flow sensitivity of cash" will be negative).
Thus, the sign and magnitude of the "cash flow sensitivity of cash" is theoretically ambiguous and will depend on whether or not firms are borrowing constrained and on whether or not increases in cash flow are accompanied by increases in productivity.
Empirical analysis is needed to determine the sign and magnitude of the "cash flow sensitivity of cash."

III. ESTIMATION MODEL AND ESTIMATION METHODS
In this section, we describe our econometric model, which is based on the theoretical considerations discussed in the previous section, as well as the estimation methods used to estimate our model.
Following Almeida, Campello, and Weisbach (2004), Khurana, Martin, and Pereira (2006), and Riddick and Whited (2009), we estimate the following baseline equation: where CHCASHA(i,t) is the ratio of the change in the stock of cash to total assets, CFA(i,t) is the ratio of cash flow to total assets, q(i,t) is Tobin's q, and SIZE(i,t) is firm size measured by total assets of firm i at time t. e(i,t) is an error term of firm i at time t.
As explained in the previous section, Almeida, Campello, and Weisbach (2004) predict that the coefficient of cash flow, a1, will be positive in the case of borrowing constrained firms but that it will be indeterminate in the case of unconstrained firms.
This prior implies that firms should increase their stocks of liquid assets in response to positive cash flow innovations if they are borrowing constrained. By contrast, unconstrained firms should not display such systematic behavior when managing their liquidity; i.e., their cash flow sensitivity of cash should not be statistically different from zero.
This prediction of a positive propensity to save out of cash flow for borrowing constrained firms is due primarily to Almeida, Campello, and Weisbach's (2004) assumption that an increase in cash flow is not accompanied by higher capital productivity. Therefore, the firm has no incentive to transform liquid assets into physical assets, and an increase in cash flow produces a pure positive income effect on saving.
Riddick and Whited (2009), on the other hand, stress the importance of the substitution effect. The substitution effect implies that the firm saves less in the form of liquid assets because it wants to shift some of its liquid assets into physical assets that have become relatively more productive.
We use two definitions of Tobin's q (q1, the ratio of market price to book value, and q2, the sum of market capitalization and debt as a ratio of total assets). However, as the results were not very sensitive to the definition of Tobin's q, we report only the results for q1.
We do the estimations with and without one-digit industry dummies, and although the results were not found to be very sensitive to the inclusion of industry dummies, we report the results with and without industry dummies.
Finally, in addition to estimating the baseline regression [equation (1)], we also tried augmenting the regression with three additional explanatory variables (capital expenditures, working capital, and short-term debt).
We estimate this equation using three estimation methods: ordinary least squares (OLS), the procedure proposed by Fama and MacBeth (1973), and the generalized method of moments (GMM). The Fama and MacBeth (1973) procedure consists of estimating the equation for each year using OLS, then pooling the yearly estimates. Since it is likely that there is measurement error in Tobin's q, we follow Riddick and Whited (2009) in using GMM to control for measurement error in Tobin's q.
We do the estimates for the full sample, for the subsamples of developed economies and developing economies, and for individual economies.
In addition, in order to gauge the impact of the global financial crisis on the saving behavior of firms, we try dividing the time period of our analysis into pre-crisis Finally, we try dividing the sample into various subsamples by firm size on the grounds that firm size will be a good proxy for borrowing constraints. In particular, we try dividing the sample into firms above and below the mean (median) and into firm size quintiles on the grounds that firms that are relatively small will be more likely to be borrowing constrained whereas firms that are relatively large will be less likely to be borrowing constrained.

IV. DATA SOURCE
In this section, we discuss the source of the data used in our analysis. firm-year observations with missing data and for which total assets, the gross capital stock, or sales are either zero or negative, selected the longest consecutive time series of data for each firm, and deleted firms with only one observation. Following Riddick and Whited (2009), we also omitted all firms whose primary Standard Industrial Classification (SIC) code is between 4900 and 4999, between 6000 and 6999, or greater than 9000 because our model is not appropriate for regulated, financial, or quasi-public firms. Finally, we also excluded outliers (defined as the top and bottom 1% of firms) and firms from economies with relatively few observations.
Appendix Table 1 shows the variable definitions and data sources for the variables used in the empirical analysis whereas Appendix Table 2 shows the summary statistics for these variables.

V. ESTIMATION RESULTS
In this section, we discuss our estimation results. Our estimation results are shown in Tables 1-9. The first three tables show estimation results by economy. Table 1 shows the OLS estimates, Table 2 shows the Fama and MacBeth (1973) estimates, and Table   3 shows the GMM estimates. Two measures of Tobin's q were used in the estimations, with the first being the ratio of market price to book value and the second being the sum of market capitalization and debt divided by total assets. The left-hand panel of Tables 1 and 2 shows the estimates using the first definition of Tobin's q, while the right-hand panel shows the estimates using the second definition of Tobin's q. Only the first measure of Tobin's q based on the ratio of market price to book value is used in Table 3.
As can be seen from Table 1, in the OLS estimates, the coefficient of greatest interest (the coefficient of cash flow) is positive and statistically significant in all of the 11 economies in our sample regardless of whether or not industry dummies are included.
As can be seen from Table 2, the Fama and MacBeth (1973) estimates are generally consistent with the OLS estimates with a positive coefficient on cash flow, but the coefficient of cash flow is statistically significant in a fewer number of economies.
The coefficient of cash flow is positive and statistically significant in six out of the 11 economies in the sample when industry dummies are included and in nine out of the 11 economies in the sample when industry dummies are not included.
As can be seen from Table 3 in the GMM estimates, the coefficient of cash flow is statistically significant in nine out of the 11 economies in the sample and is positive in all cases in which it is statistically significant.
Turning to the coefficient of Tobin's q, in the OLS estimates, it is statistically significant in seven (six) out of the 11 economies in the sample (positive and significant in five (five) economies, and negative and significant in two (one) economies when industry dummies are included (omitted). In the Fama and MacBeth (1973) estimates, it is not statistically significant in any of the 11 economies in the sample regardless of whether or not industry dummies are included. In the GMM estimates, the coefficient of Tobin's q is statistically significant in only four out of the 11 economies in the sample and is positive in three out of the four economies in which it is statistically significant.
We also obtained the GMM results for the full sample of economies, the developed economy sample, and the developing economy sample, and the results are shown in Tables 4, 5, and 6, respectively. As can be seen from these tables As for the impact of Tobin's q, there is some evidence that its impact on the change in the stock of cash is positive but the results are not very clear-cut.
Turning to the results for the pre-and post-global financial crisis periods, the coefficient of cash flow is almost always higher during the pre-crisis period than during the post-crisis period, which is reasonable since the greater pessimism about future prospects caused by the global financial crisis presumably reduced the demand for cash holdings in preparation for the sudden and unexpected appearance of profitable investment opportunities. The only exception to this pattern is in the case of the baseline regression results for the developing economy sample, but the aforementioned pattern holds in the case of the augmented regression results for this sample, which are presumably more reliable than the baseline regression results due to the inclusion of more explanatory variables.
We now divide the sample into subsamples according to firm size (asset size) in order to test for the possibility of the differential sensitivity of the change in the stock of cash to the cash flow variable by asset size. In this exercise, we use the asset size of firms to divide the sample of firms in each economy into five groups (quintiles). The grouping is based on the average or median asset size of firms in each economy, and thus, for firms in an economy in which average (median) asset size is relatively large such as the PRC, Japan, or the Republic of Korea, some of the firms in the bottom quintile might be larger than those in the highest quintile in other economies such as Viet Nam where average asset size is relatively small.
The results are shown in Tables 7, 8, (2009) predicted. This is true even when we augmented the specification by adding capital expenditures, working capital, and short-term debt and estimated the equation using GMM. This finding constitutes strong support for the income effect posited by Almeida, Campello, and Weisbach (2004), and there is no evidence of a shift from cash flow to physical assets.
Turning to patterns by firm size, as explained earlier, Almeida, Campello, and Weisbach (2004) predict that the coefficient of cash flow, a1, will be positive in the case of borrowing constrained firms but that it will be indeterminate in the case of unconstrained firms. Since smaller firms are more likely to be borrowing constrained, we would expect the coefficient of cash flow, a1, to be larger and more highly significant in the case of smaller firms, and our results are consistent with this expectation.
As the results for developed economies in Table 5 show, the coefficient of cash flow is positive and significant in the case of the lowest firm size quintile but is not statistically significant in any other quintile. Moreover, as the results for developing economies in Table 6 show, the coefficient of cash flow is positive and statistically significant in the case of all firm sizes, but its significant level and absolute magnitude decline with firm size in the case of the baseline regressions (although the coefficient of cash flow is not significant for any firm size in the case of the augmented regressions).
Thus, the results for both developed and developing economies are consistent with our prior that smaller firms are more likely to be borrowing constrained and that they would be expected to have a stronger tendency to save when cash flow is high.

VI. SUMMARY AND CONCLUSIONS
In this paper, we analyzed the determinants of corporate saving in the form of changes in the stock of cash for 11 Asian economies using firm-level data from the Oriana Database for the 2002-2011 period. We found some evidence that cash flow has a positive impact on the change in the stock of cash, which suggests that Asian firms are borrowing constrained and that they save more when their cash flow increases so that they will be able to finance future investments. Moreover, we found in the developed economy sample that, as expected, cash flow has a positive impact on the change in the stock of cash only in the case of the smallest firms, which are more likely to be borrowing constrained, and find in the developing economy sample that, as expected, the positive impact of cash flow on the change in the stock of cash declines with firm size. In addition, we find that the cash flow sensitivity of cash declined after the global financial crisis. Finally, we found some evidence that Tobin's q has a positive impact on the change in the stock of cash.             Robust standard errors are reported below estimates in parentheses * p<0.05, ** p<0.01, *** p<0.001 Instruments: Two lags of tangible fixed assets, lagged working capital, lagged short-term debt , twice-lagged sales growth, and country dummies.   Robust standard errors are reported below estimates in parentheses * p<0.05, ** p<0.01, *** p<0.001

FULL SAMPLE
Instruments: Two lags of tangible fixed assets, lagged working capital, lagged short-term debt , twice-lagged sales growth, and country dummies.   Robust standard errors are reported below estimates in parentheses * p<0.05, ** p<0.01, *** p<0.001

POOLED DEVELOPED COUNTRIES
Instruments: Two lags of tangible fixed assets, lagged working capital, lagged short-term debt , twice-lagged sales growth, and country dummies.