“Policy initiatives and firms’ access to external finance: Evidence from a panel of emerging Asian economies

This paper analyses the impact of policy initiatives co-ordinated by Asian national governments on (cid:133)rms(cid:146)composition of external (cid:133)nance. Using a unique (cid:133)rm-level database of eight Asian countries- Hong Kong SAR, Indonesia, Korea, Malaysia, Philippines, Singapore, Taiwan and Thailand over the period of 1996-2012 and a di⁄erence-in-di⁄erences approach, the results show a signi(cid:133)cant impact of policy initiatives on (cid:133)rms(cid:146)choice to external (cid:133)nance. We (cid:133)nd that (cid:133)rms increased their uptake of long-term debt, while decreased their short-term debt. We also document that less risky and more pro(cid:133)table (cid:133)rms are more signi(cid:133)cantly a⁄ected by the policy change than riskier and less pro(cid:133)table (cid:133)rms. Finally, we show that the improved access to external (cid:133)nance after the policy initiative helped (cid:133)rms to raise their investment spending. more detailed review of the (cid:133)nancial systems in Asian is presented by and They highlighted that bond markets expanded by almost 57% in East Asia, 345% in China and 66% in India, during the 2000s relative to the 1990s. percentage point e⁄ects by dividing the coe¢ cient value (marginal e⁄ect) with the predicted probability of the model. We show that the introduction of the policy in the treated group led to a reduction in (cid:133)rms(cid:146)access to short-term debt by 4.87% and an increase in (cid:133)rms(cid:146)access to long-term debt by 11.96%. This on the suggest their access to long-term and reduced their short-term debts after the introduction of the results show (cid:133)rms to their corporate bond issuances and reduce their bank (cid:133)nance more compared to their (cid:133)nancially constrained counterparts. we take into account the in(cid:135)uence of (cid:133)rms(cid:146)external (cid:133)nance on investment spending in the post-policy period. The results show that increased access to credit for (cid:133)rms in the form of total long-term debt had a positive impact on (cid:133)rms(cid:146)investment spending. Finally, we that with respect to long-term debt ratio, unconstrained (cid:133)rms are able to increase their post-policy investment spending much more compared to constrained (cid:133)rms due to their increased access to long-term debts after policy.


Introduction
The 1997-98 Asian …nancial crisis prompted policymakers in Asia and the Paci…c to initiate a series of reforms aimed at the development and strengthening of the regional bond markets. The most prominent initiative of Asian Bond Funds (ABF and ABF-2) was undertaken in two waves in 2003 and 2005. The main role of this policy intervention was to broaden investor participation and improve market infrastructure in order to expand the domestic bond markets. As post-ABF data become widely available and their Latin American counterparts seek to adopt similar programmes, it is important to examine the extent to which this program has been successful in meeting the above objectives. The present paper provides new evidence on corporate …nancial choices after the introduction of policy initiatives in Asia. More speci…cally, we ask whether these initiatives have led to better/improved access of …rms to external …nance compared to …rms that operate in an economy that did not participate in this policy intervention.
There is evidence that the ABF encouraged expansion and liquidity in the Asian domestic bond markets by implementing several market reforms such as liberalising foreign exchange administration rules, tax reforms to exempt withholding tax of non-resident investors, improving regulatory framework for exchange traded funds, strengthening domestic market infrastructure and decreasing cross-border settlement risk, creating transparent and credible bond indices (Packer and Remolona, 2012). These reforms helped in expanding the domestic bond markets and their issuer base. The share of local currency bond markets increased substantially from 42.8% since the launch of the ABMI in 2003 to 54.5% in Q3 2008, thus improving market liquidity in the Asian markets (Spiegel, 2012) 1 .
Comparing the issuance of long-term debt of participating …rms in our sample with non-participating …rms reveals a large wedge between the two groups: USD 66.89 billion for the former versus 2.91 billion for the latter. Even when making a before-after comparison, we can see that the average long-term debt of participating …rms increased from USD 59.25 billion before the ABF to 72.44 billion after the ABF. However, it is not clear how much of this change stems from the policy change itself and how much is due from improvements in the general economic climate. A di¤erence-in-di¤erences model will tease out the regional bond development and policy in ‡uences. In order to separate the e¤ects of this regional development from the e¤ect of regional policy initiatives, we refer to Taiwan as a control. Taiwan saw similar development in its national bond market and is comparable to the other Asian economies, but it did not participate in the ABF, ABF2 or ABMI initiatives.
The identifying assumption for the research design is that economies that participated in the policy initiative and those that did not would have trended similarly in the absence of the policy change. The parallel trends assumption is supported by the institutional background of the Asian bond initiatives as well as graphical evidence 2 . Figure  Panel B, which shows the evolution of corporate bond market, displays a similar pattern.
It shows a widening gap between the control and treated group after the introduction of the policy initiatives. While, panel C shows that the control group (Taiwan) showed similar growth patterns over the years with majority of East Asian economies in terms of bond market development 3 . Figure 2 graphs the development of the banking sector in the treated and control group over the same period. The graph displays the growing trend of the banking sector in the treated group after the Asian …nancial crisis, while the growth in the control group almost remains constant throughout the period. The graphs described above con…rm the parallel trends assumption in our data which suggests that in the absence of the policy change the two groups would have continued to track each other.
Much of the empirical investigation of …rms' external …nance considers how exogenous shocks a¤ect …rms' …nancing choices. Bris et al. (2014) shows that non-…nancial …rms from euro countries with previously weak currencies considerably increased their debt …nancing after the introduction of the Euro. Leary (2009) highlights the impact of a market for certi…cates of deposit which led to increase in bank credit and …rms' leverage. Su… (2009) studies the impact of syndicated loan ratings introduced in 1995 on borrowing of …rms that obtained a rating, …nding that …rms that are able to obtain a rating are able to use more debt, which also increases …rms'asset growth, cash acquisitions, and investment in working capital. Further, Lemmon and Roberts (2010) show that the collapse of Drexel Burnham Lambert had a negative e¤ect on …rms'…nance and investments, especially for those which were using high-yield debt …nancing. In the Asian context, recent work by Mizen and Tsoukas (2014) demonstrates that regional initiatives have been an important step towards greater bond issuance by …rms in Asia, mostly by fostering market deepening and improving liquidity.
But it is not clear whether the demand for all types of external …nance will increase proportionately. Some components of external …nance may prove more attractive than others, and the demand for bank borrowing for example, may exceed demand for credit obtained through the sale of marketable debt or equity. Equally, short-term debt may be replaced by longer-term debt if short-term interest rates fall below the rates that are expected to prevail in the longer run. In addition, some types of …rms may …nd that the credit constraints that they face on the supply side, from the markets and …nancial intermediaries, are relaxed less quickly than for other types of …rms. Firms that were previously only able to obtain credit from banks, perhaps because they were too small to be listed on the stock exchange or to issue their own commercial paper or bonds, …nd that other forms of credit become more readily available.
In this paper we look at the …nancial health of the …rm, re ‡ected in the quality of its balance sheet. Then we consider di¤erent types of external …nance such as short-term or long-term debt and we construct ratios that measure …rms' choice of external …nance. Finally, we observe a unique policy experiment, namely the ABF initiative, which will be used to identify the e¤ects of the policy change on …rms'composition of external …nance. Our empirical work is based on an assessment of the policy initiative on …rms' access to external …nance using an unbalanced panel of 7,436 Asian listed …rms between 1996-2012. We merge data from di¤erent sources including Bondware, Bloomberg, Standard and Poor's Compustat Global database, Global Financial database and IMD World Competitiveness Yearbook.
Our paper contributes to the existing literature in three important ways. First, we analyse the role of the Asian Bond Funds (ABF and ABF2) and the Asian Bond Market Initiative (ABMI) in the composition of external …nance. While previous studies identi…ed that the regional initiatives in Asia had a greater and positive impact on …rms'decisions to issue bonds in Asia (see Mizen and Tsoukas, 2014), we go one step further by examining the impact of these initiatives on …rms'external …nance using the di¤erence-in-di¤erences method.
Second, we exploit …rm-level heterogeneity by considering whether …rms that face …nancial constraints may be more or less likely to alter their composition of external …nance. In doing so, we employ characteristics such as …rms'pro…tability and coverage ratio as measures of …nancial constraints. These characteristics are likely to be critically important in in ‡uencing …rms'access to …nancial markets and we intend to explore how the interplay between changes in …nancial health and the introduction of policy initiatives, a¤ect the choice of external …nance for more and less constrained …rms.
Finally, we build on the extant literature of …rms'investment spending (Fazzari et al., 1988;Almeida and Campello, 2007), by considering whether the policy intervention by the Asian regional governments has impacted on …rms'investment spending. Hence, we explore the relationship between external …nance and …rm's investment spending before and after the policy initiative. We also allow for the fact that …rms of di¤erent riskiness, with varying levels of pro…ts might respond to the policy initiative disproportionately.
The rest of the paper is organized as follows. Section two provides a brief review of the relevant literature. In section three, we present the data used in our empirical analysis along with summary statistics. We describe the econometric modelling strategy in section four, and we report the econometric results in section …ve. In section six we check the robustness of our …ndings and we provide concluding remarks in section seven.

Background literature
In the past, Asian bond markets have been identi…ed as under-developed and illiquid. Minimum e¢ cient scale, corruption and low level of bureaucracy, poor accounting standards (Eichengreen and Luengnaruemitchai, 2006), capital controls, taxation, limited availability of hedging instruments (Takeuchi, 2006) were some of the obstacles in developing the bond markets in Asia. However, since the 1990s many emerging countries in Asia have taken important steps to enhance their …nancial markets. Especially after the Asian crisis, the need to develop sound and more liquid bond markets to prevent further capital account crisis was realised and regional governments started contributing towards the development of local bond markets in Asia. interest in Asian bonds. In our study, we will make use of this natural experiment to evaluate its in ‡uence on …rms'choice of external …nance.
A more detailed review of the …nancial systems in Asian countries is presented by Didier and Schmukler (2014). They highlighted that bond markets expanded by almost 57% in East Asia, 345% in China and 66% in India, during the 2000s relative to the 1990s.
Despite of substantial growth between 2000 and 2009, private bond markets including corporate and …nancial institutions in Asian economies remain relatively small compared to G-7 economies. In contrast, the developments in the Asian bond market are still the highest among other developing countries. For instance, the bond markets in Eastern Europe and Latin America grew respectively at 28% and 32% of GDP, exceptionally lower than 56% in East Asia. The banking system in the East Asian economies increased by 47% between 1980-89 and 2000-2009, while in Eastern Europe, G-7 economies and Latin America increased by 25%, 20% and 5%, respectively during the same period.
The structure of private credit and public credit have changed considerably across the world over the past two decades. In East Asian economies, private sector lending has increased from 44% to 72% of GDP, compared to 50% in the 1980s to 98% in the 2000s in other advanced economies. While, public sector lending accounts 10% and 13% of total claims by the banking sector in G-7 and East Asian economies during the 2000s. Thus, these …nancial trends overall highlight the fact that Asian economies are more developed compared to Eastern Europe and Latin America, due to rapid …nancial improvements in the region during the 2000s.
Moving to the literature which attempts to identify the main determinants of …rms' choice of external …nance, we note that there is a variety of …rm-level and macroeconomic factors that in ‡uence …rm …nancing. With respect to the …rm-level characteristics, size, age, collateral, pro…tability and riskiness are highlighted as important factors a¤ecting access to external …nance by …rms (see Bougheas et al., 2006). Business group a¢ liation (Shin and Park, 1999) and availability of credit information (Tang, 2009) are other factors a¤ecting …rms' choice of capital structure. At the macroeconomic level, monetary policy shocks (Kashyap et al., 1993), institutional di¤erences and development of …nan-cial markets (Demirgüç-Kunt and Maksimovic, 2002;Fan et al., 2010) and higher costs and barriers to entry (Cetorelli and Strahan, 2006) are typically highlighted as important variables in ‡uencing …rms'access to external …nance.
The studies discussed above provide a useful background to explore the linkage between …rms'access to external …nance and Asian bond market initiatives. In addition, we also try to analyse the impact of debt availability on …rms'investment spending in the post-policy period. In the following sections, we explain our data and empirical strategy.

The baseline model
To examine the impact of bond market policy initiatives on …rms'access to external …nance, we employ di¤erence-in-di¤erences methods 5 . More speci…cally, we evaluate the impact of the policy intervention by comparing di¤erences in external …nance composition before and after the policy change for …rms in seven Asian economies that participated in this initiative (treatment) namely-Hong Kong SAR, Indonesia, Korea, Malaysia, Philippines, Singapore and Thailand. We then compare these di¤erences with comparable di¤erences from a sample of …rms in Taiwan (control), which did not participate in the policy initiative but faced similar trajectory in …rms'…nancing activities.
The dependent variables capture measures of external …nance and are based on the ratios of short-term debt to total debt (Bougheas et al., 2006) and long-term debt to total assets (Demirgüç-Kunt and Maksimovic, 1999) 6 . The former ratio refers to access to bank …nance versus market …nance, while the latter ratio is more related to access to bond …nancing, compared to total assets held by the …rms. These ratios help to remove demand-side in ‡uences as increase in the demand of credit is likely to a¤ect both 5 Several recent studies use this methodology to estimate the impact of minimum wages on employment (Card and Krueger, 1994;Leigh, 2003), the e¤ect of grants and subsidies on exporting activity (Görg et al., 2008) and the role of the …nancial crisis in corporate investment (Duchin et al., 2010). 6 Short-term debt is made up of the sum of bank overdrafts, short-term group and director loans, hire purchase, leasing and other short-term loans, but is predominantly bank …nance. Long-term debt is made of bonds, mortgages, loans and similar debt, which represents debt obligations due more than one year from the company's balance sheet date or due after the current operating cycle. numerator and denominator of the ratio, leaving the ratio unchanged (Bougheas et al., 2006). We estimate a baseline model of the following kind: (1) where i = 1, 2, . . . ., N refers to the cross-section of units (…rms in this case), in country j T A ijt are the ratios of short-term debt to total debt and long-term debt to total assets, respectively. T reat j is a country dummy which takes value one if a country participates in the policy initiative and zero otherwise. Af ter t is a time dummy which takes value one for the period 2005-2012 and zero otherwise 7 . The policy e¤ect is given by the coe¢ cient of the interaction term, T reat j Af ter t . The coe¢ cient on the interaction term measures the di¤erence in the choice of external …nance between those …rms that experience an anticipated policy change and those that do not. If the policy initiative has a crucial e¤ect on access to …nance with a considerable di¤erence between the control and treated group, then it should have a signi…cant coe¢ cient value.
The models are estimated using di¤erence-in-di¤erences with …rm …xed e¤ects to control for unobserved heterogeneity at the …rm-level. Country dummies are included in the model to control for country-level di¤erences, while time dummies interacted with industry dummies are included in order to control for all time-varying demand shocks at the industry level (Brown and Petersen, 2009). We cluster standard errors at the …rm level as the observations over time might be correlated within …rms. Finally, X is a vector which includes other explanatory factors at both …rm and country-level and e ijt are the disturbance terms.
In vector X we use a number of …rm-level and macro-economic variables which have been found to in ‡uence …rms'choice of external …nance. To begin with the …rm-speci…c 7 By observing the treatment after 2005, we are mainly focusing on the impact of ABF-2 introduced at the end of 2004. ABF-2 initiative di¤ers from the others as it involves the actual creation of local currency bond funds. The earlier ABF initiative had limited itself to dollar-denominated issues that are traded mostly in more developed international bond markets. ABF was important because it a¤orded the EMEAP central banks an opportunity to work together for building trust in order to foster cooperation and further develop …nancial markets in the region (Ma and Remolona, 2006). characteristics, previous literature postulates that size is an important determinant of external …nancing. Firm size is measured as the natural logarithm of total assets (Mizen and Tsoukas, 2014). Larger …rms have better access to external …nance as they are less …nancially constrained, while smaller …rms are more dependent on short-term bank …nancing (Bougheas et al., 2006).
Firms'liquidity is measured by the ratio of current assets to current liabilities. According to Ozkan (2001), liquidity of the …rms is likely to have a mixed impact on access to external …nancing. Higher liquidity might encourage …rms to have higher debt ratios due to increased ability to meet short-term obligations, implying a positive relationship between liquidity and external …nance. While, …rms with higher liquidity might also reduce their debt access exerting a negative impact on external …nance. Following Bougheas et al. (2006), gearing is measured by total liabilities to shareholder's equity. González et al. (2007) show that more leveraged …rms have fewer requirements of external …nancing. 8 Following Mizen et al. (2012), the expansion rate of …rms is measured by the ratio of investments to total assets. According to Datta et al. (2000), faster growing …rms are more likely to issue bonds compared to …rms with less opportunities for expansion. Also, …rms with higher expansion rate are likely to undertake bond issuance earlier (Hale and Santos, 2008). Firm's operating cycle is calculated as the ratio of net sales to net …xed assets. Firms with higher operating cycle, depend more on short-term debt to …nance the sales (Demirgüç-Kunt and Maksimovic, 1999). Finally, cash ‡ow is measured by the ratio of earnings before extraordinary items plus depreciation and amortization to total assets (Almeida and Campello, 2010). Firms with higher levels of cash ‡ow or higher cash surplus are expected to reduce their leverage (Demirgüç-Kunt and Maksimovic, 1999). In addition to …rm-level variables, we also control for other macro-economic factors in vector X such as GDP growth rate, legal regulation and balance of trade 9 . Among these factors, GDP growth rate and balance of trade (scaled by GDP) are measures of general economic development. Better economic conditions might encourage …rms to shift towards non-debt liabilities showing a negative e¤ect on external …nance. Firms in countries with higher levels of legal regulation are more likely to rely on external …nancing due to reduced information asymmetries resulting in higher growth (La Porta et al., 1998) and better working of …nancial contracts (Demirgüç-Kunt and Maksimovic, 1999).

Access to external …nance and investment after the policy change
In this section, we explore the in ‡uence of external …nance on …rm's investment spending in the post-policy period 10 . One of the objectives of the Asian bond market initiatives is to provide alternative sources of …nancing for private and public investments to enterprises (Kawai, 2010). It is particularly interesting to examine the impact of long-term debt issued on …rm's investment spending as the Asian Bond Fund initiatives are expected to expand long-term debt issuance. We argue that this is likely to have a positive e¤ect on …rms'investment spending. To test this hypothesis, the dependent variable of …rm's investment spending (Inv) is measured as the ratio of annual capital expenditure to total assets (Duchin et al., 2010). The models are estimated as follows: where Lev ijt refers to the measures of external …nance i.e. both short-term and longterm debt ratios, CF measures …rm's cash ‡ow and Q controls for …rm's investment opportunities. The main variable of interest is the interaction term between leverage and the DD coe¢ cient which is given by T reat j Af ter t Lev ijt , captures the impact of postpolicy access to external …nance on …rms'investment expenses for the treated group. Due to data unavailability on market value of assets (e.g. number of shares outstanding and stock price) in Global Compustat we are unable to construct Tobin's Q (see also Baum et al., 2011), but investment opportunities are controlled for in two ways. First, following Konings et al. (2003) and Bakucs et al. (2009), sales growth is used as a proxy for Tobin's Q. Second, time dummies interacted with industry dummies in all the speci…cations are used as an indirect way of controlling for investment opportunities (see Guariglia et al., 2012).

Accounting for …rm-level heterogeneity
Intuitively, not all …rms are expected to bene…t equally from the above mentioned policy initiatives. Fazzari et al. (1988) highlight the importance of di¤erences across …rms in relation with …nancial constraints originating from the imperfections of the capital market. Due to asymmetric information …rms facing higher costs of external …nance are likely to be more …nancially constrained. Bris et al. (2014) …nd that larger …rms in the Euro area bene…ted the most from …nancial integration. Consistent with this result, Gozzi et al. (2010) …nd that larger …rms have better …nancing from international capital markets. Stiebale (2011) further stressed that …nancially constrained …rms face di¢ culties in obtaining external …nance. We also argue that …rms that face …nancial constraints might be less well positioned to take advantage of the policy initiatives in Asia since these …rms are more susceptible to information asymmetry e¤ects. It is well known that there is little public information available for …nancially constrained …rms and it is di¢ cult for …nancial institutions to gather this information. Obtaining external …nance is therefore likely to be particularly di¢ cult and/or costly for them. We hypothesise, therefore, that …nancially unconstrained …rms are more likely to reap the bene…ts of a policy change.
To test this hypothesis, we divide our …rms into constrained and unconstrained groups using two main criteria: pro…ts and coverage ratio. The former classi…cation scheme is measured by the ratio of earnings before interest and taxes relative to total assets (Baker and Wurgler, 2002). Evidence shows that less pro…table …rms are more …nancially con-strained (Livdan et al., 2009). Coverage ratio is measured as earnings before interest and taxes over total debt which measures project quality (Mizen and Tsoukas, 2012). Hence, this classi…cation scheme captures …rms'creditworthiness. 11 As the policy initiative might be related to unobserved within-…rm changes, …rms are divided into constrained and unconstrained categories using the pre-policy period of 1996-2004 12 . Firms are classi…ed as constrained if their pro…ts and coverage ratio are below the 50th percentile of the distribution in the pre-policy period.
Further, we investigate the di¤erential impact of the in ‡uence of external …nance on …rms'investment spending across constrained and unconstrained …rms. Theory predicts that …rms with …nancial frictions accompanied with negative shocks to external …nance might lack su¢ cient …nancial slack to fund pro…table investment opportunities internally (Holmstrom and Tirole, 1997). These e¤ects are stronger for constrained …rms that face higher costs in raising external capital (Duchin et al., 2010). Bondware is used to identify corporate bonds issued in international markets and to 11 Interest coverage was used by Guariglia (1999) as an indicator of …nancial constraints to identify di¤erences in inventory investment. 12 A …rm is classi…ed as constrained or unconstrained in the post-policy period of 2006-2012 using values of 2004, one year prior to the onset of the policy as …rm variables are likely to be endogenous to the choices made by …rms. 13 We do not take into consideration China due to its disparities with other East Asian economies in terms of growth, capital account convertibility and restricted …nancial markets. Corporate savings in China are higher due to the domination of state-owned banks and restricted equity market, which favours the large …rms by improving their retained earnings and pro…tability (Lin, 2009). In addition, the repressed …nancial system in China provides cheap capital (lower interest rates) which again favour large …rms (Prasad, 2009). assemble data relating to issue date, maturity date, outstanding amount and currency 14 .
We use Bloomberg to categorise similar data for …rms that issue bonds in the Asian domestic markets. The data thus, covers bond issues of …rms denominated in local currency as well as in foreign currency, mainly US dollars.
Balance sheet and pro…t (loss) data are taken from Compustat Global, which provides market information on active and inactive companies throughout the world. Our initial sample included a total of 71,792 annual observations on 7,436 companies. Finally, the data for economic factors such as GDP growth rate, balance of trade (scaled by GDP) and legal regulation are drawn from the Global Financial database and IMD WCY database.  Table 1 shows summary statistics for all the variables, distinguishing between control and treated groups, as well as before and after the introduction of the policy initiative.

Summary statistics
We report values for the whole sample (column 1); treated and control groups (columns 2 and 3); before and after the policy initiative (columns 5 and 6). We also report pvalues for the test of equality of means (columns 4 and 7). From columns 2 and 3 we …nd signi…cant di¤erences in the short and long-term debt issued across the two groups of economies. Further, columns 5 and 6 show the debt levels for pre-and postpolicy periods. Regarding the short-term debt we are unable to observe any signi…cant di¤erences between the two time periods. On the other hand, long-term debt values show considerable improvement after the introduction of the policy and this di¤erence is signi…cant at the 5% level. With respect to other variables, we …nd that …rms are larger, have higher levels of gearing, lower liquidity and expansion rate and higher investment spending in the treated group compared to the control group. As for country-speci…c variables, the treated group displays higher GDP growth, lower legal regulation and higher balance of trade compared to the control group. Moving to columns 5 and 6, all variables show signi…cant di¤erences before and after the policy with GDP growth being the only exception. Introduction of the policy, on average, helped in improving …rm-level factors such as …rm size, liquidity, expansion rate and operating cycle of …rms.
Overall, two points can be highlighted from these preliminary statistics. First, there is a noticeable di¤erence between the control and treated group in terms of both short and long-term debt issuances. Second, there is an increase in the level of long-term debt after the introduction of the policy initiatives. In addition, there is a notable improvement in the performance of …rms highlighted by the …rm-level characteristics after the policy initiative. In the following sections, a formal regression analysis framework tests the role of the policy initiative in …rms'access to external …nance. Table 2 reports the results for the baseline model. The main variable of interest is the DD estimate, T reat j Af ter t , which signi…es the impact of the policy initiatives on the treated group. We …nd evidence that the introduction of the ABF initiative had a important e¤ect on the …rms' choice of external …nance. Firms' decreased their shortterm debt, while increased their uptake of long-term debt. To ascertain the magnitude of the DD coe¢ cient, we calculate percentage point e¤ects by dividing the coe¢ cient value (marginal e¤ect) with the predicted probability of the model. We show that the introduction of the policy in the treated group led to a reduction in …rms'access to shortterm debt by 4.87% and an increase in …rms'access to long-term debt by 11.96%. This …nding highlights the fact that policy initiative was bene…cial for the …rms in the treated group as it helped them to increase their access to long-term debt while reducing their short-term debt …nance, compared to the …rms in the control group. In other words, …rms issued long-term debt that they would not have done in the absence of the policy change.

The baseline model
Our …ndings lend support to evidence presented in Mizen and Tsoukas (2014), who show that the policy initiatives of ABF, ABF-2 and ABMI had a signi…cant e¤ect on a …rm's decision of bond issuance. We also support the …ndings of Shim (2012)  to 2011 compared to other emerging markets, even during the global …nancial crisis. One of the factors a¤ecting the rapid growth in corporate bond issuances is the functioning of credit rating agencies, which are established as a part of the government initiatives (Shim, 2012).
Focusing on …rm-level factors, a signi…cant coe¢ cient on …rm's size shows that larger …rms have greater access to long-term debt and reduce their short-term debt. This con…rms the …nding by Bougheas et al. (2006) that size is an important determinant of access to bank and market debt. Liquidity of …rms shows a negative impact on short-term debt while, a positive impact on long-term debt suggesting that …rms with higher liquidity are likely to raise more long-term debt and reduce short-term debt. Ozkan (2001) show that higher liquidity of …rms improve access to external …nance. Moving to gearing of …rms, it enters with the expected negative and positive signs on short and long-term debt ratios, respectively. This result is in line with González et al. (2007) for short-term debt as they con…rm that more leveraged …rms have less desire for external …nancing. On the other hand, …rms with higher leverage are likely to issue more corporate bonds (Mizen and Tsoukas, 2014), resulting in an increase in access to long-term debt.
Expansion rate measured by investments to assets ratio shows a negative and positive e¤ect on short and long-term debt ratios, respectively. This indicates that …rms with higher investments are more likely to opt for long-term debt issuance. Operating cycle measured by sales to assets ratio attains a negative coe¢ cient on long-term debt ratio while a positive coe¢ cient on short-term debt ratio. This con…rms that …rms depend more on short-term debt rather than long-term debt to …nance their increasing sales.
Cash ‡ow enters with a negative coe¢ cient on both long and short-term debt ratios which show that …rms with substantial cash ‡ow require less in terms of external …nance.
Country-speci…c determinants include GDP growth, legal regulation and balance of trade. GDP growth is generally insigni…cant, but the balance of trade shows a positive e¤ect on short-term debt and an insigni…cant e¤ect on long-term debt. Improvement in the balance of trade, an indicator of economic health of a country, increases access to external …nancing in the form of short-term debt. Finally, legal regulation registers a positive e¤ect on long-term debt ratio and a negative e¤ect on short-term debt ratio.
This implies that with an improvement in country's legal framework, …rms are more likely to increase their long-term debt issuance rather than the short-term debt exposure.

Accounting for …rm-level heterogeneity
In this section, we explore the link between the policy shift and …rms' …nancing while taking into account …rm-level heterogeneity. The results are reported in Table   3. Columns 1-2 provide results for …rms with low and high pro…ts, followed by low and high coverage ratios in columns 3-4 for short-term debt ratio. Similarly, results for the long-term debt ratio for di¤erent classi…cation schemes are reported in columns 5-8.
The estimation results provide evidence that constrained …rms were mainly una¤ected by the policy change. On the other hand, unconstrained …rms reduced their access to short-term debt and increased their access to long-term debt after the policy initiative.
In sum, we …nd that the policy initiative had an important impact on the choice of external …nance, though its proportional e¤ect was very heterogenous across constrained and unconstrained …rms. One explanation for the above …nding is related to the fact that that …nancially constrained …rms typically face higher external …nancing costs (see Fazzari et al., 1988). Hence, they are less likely to have access to external …nance as they face higher agency costs of borrowing from …nancial markets when compared with the cost of internal …nancing (Bernanke and Gertler, 1995).
In economic terms, after the introduction of the policy, unconstrained …rms in the treated group reduced their access to short-term debt by around 9.45%-9.51%, while they increased their access to long-term debt by 18.25%-26.84%, compared to unconstrained …rms in the control group. The test of equality for constrained and unconstrained …rms also shows a signi…cant di¤erence at 5% level for both short and long-term debt ratios. Thus, unconstrained …rms are able to reap more readily the bene…ts from the policy change, compared to their constrained counterparts.

The impact on investment spending
This section takes into account the role of …rms' …nancing position in in ‡uencing the impact of the policy initiative on investment spending. Table 4 reports the results of post-policy …rms' investment spending for di¤erent measures of leverage 15 . Column 1 summarises the results for short-term debt to total debt and in column 2 leverage is measured as long-term debt to total assets. We …nd that the interaction term of leverage and DD is negative for …rms' investment spending in column 1 and positive in column 2. These point estimates, which are statistically signi…cant at the 1% level, indicate that as …rms'access to short-term debt was reduced after the policy implementation, …rms curtailed their investment spending using short-term debt. On the contrary, …rms'increased access to total long-term debt after the policy helped them to spend more of long-term debt on their investment outlay.
The magnitude of the interacted coe¢ cients suggests that after the implementation of policy, …rms reduced their investment spending using short-term debt by 10.04%, while increased their investment spending using long-term debt by 67.65%.
With respect to other control variables, both sales growth and cash ‡ow have a positive and signi…cant coe¢ cient for investments in almost all the columns. This result is again in line with the empirical studies such as Fazzari et al. (1988) and Wei and Zhang (2008) which show that …rms' cash ‡ow per unit of capital is positively related to the rate of investment per unit of capital, even when a measure of Q is included as an explanatory variable of investment.
Overall, the results show that the growth of Asian domestic bond markets has helped …rms to …nance their investments by increasing their access to long-term debt. Levinger et al. (2014) show that the strong growth in Asia's corporate bond markets has made funds available for investment and expansion in recent years along with deepening of capital markets and diversi…cation of …nancing sources.

Accounting for …rm-level heterogeneity
Next, we explore the link between the policy change and …rms'…nancing while categorising …rms into …nancially constrained and unconstrained. Table 5 reports the results of the post-policy investment spending and leverage for constrained and unconstrained …rms. The results indicate that the policy initiative did not have any signi…cant impact on constrained …rms'investment spending. On the contrary, unconstrained …rms reduced their investment spending using short-term debt and increased their investment spending using long-term debt after the introduction of the policy. The economic interpretation is even more interesting since it suggests that unconstrained …rms reduced their investment spending using short-term debt by 9.78%-12.67%, while they increased their investment spending using long-term debt by 88.50%-89.60%. However, the test of equality does not show a signi…cant di¤erence between the two groups for short-term debt ratio but it shows a signi…cant di¤erence between the groups at the 5% level for long-term debt ratio. In sum, we …nd once again that it is the unconstrained group of …rms in the treated group that bene…ted the most from the policy initiative in comparison to the …rms in the control group.
6 Robustness tests 6.1 Propensity score matching To check the validity of our treated and control groups, we employ a propensity score matching technique. We use one to one matching technique of our …rms without replacement. This means that once an untreated …rm has been selected to be matched to a given treated …rm, that untreated …rm is no longer available for consideration as a potential match for subsequent treated …rms. Hence, each untreated …rm is included in at most one matched set. Matching without replacement increases the e¢ ciency of matching, compared to matching with replacement. However, despite theoretical di¤erences several studies have provided evidence that the number of matches and the choice of matching with or without replacement has a minimum e¤ect on treatment e¤ect's bias and e¢ ciency (Stuart, 2010). Matching is done using non-categorical variables such as …rm size, liquidity, leverage, expansion rate, operating cycle and cash ‡ow, using caliper 0.001 (Yörük, 2008) 16 .
The results of the main variables of interest are reported in Table 6. The results in panel A con…rm the signi…cant and positive impact of the policy initiative on …rms'access to long-term debt ratio and the negative impact on access to short-term debt ratio. Panel B validates that the policy initiative helped unconstrained …rms to increase their access to long-term debt and reduce their short-term …nance. Panel C shows that …rms reduced their investment spending using short-term debt, while they increased their investment spending using long-term debt after the policy was introduced. Finally, in panel D we …nd that the relationship between external …nance and investment spending is stronger for unconstrained …rms compared to their constrained counterparts. Thus, we con…rm that our results are robust to a matching technique, which also vindicates the validity of the control and treatment groups in our main models.

Using the ABF index as a measure of treatment
To further corroborate the accurate identi…cation of the policy initiative on the treated group, we use an index as a measure of the treatment. Speci…cally, we use the Markit iBoxx ABF index which is designed to re ‡ect the performance of the local currency denominated sovereign and quasi sovereign debt from 8 Asian countries/territories. The 16 Although not reported here, we also use caliper as 0.0001 for the propensity score matching method. In every procedure with calipers 0.001 and 0.0001, the propensity score and the coe¢ cient estimate of almost all the control variables are statistically indi¤erent between the treated and control group. index gives a broad coverage of the sovereign and sub-sovereign bond universe of the treated countries whilst upholding minimum standards of investability and liquidity 17 .
The results shown in Table 7 con…rm that our main …ndings are upheld. We continue to observe that …rms reduced their access to short-term debt, while increasing their access to long-term debt in the post-policy period. Further, we …nd that …rms reduced their investment spending using short-term debt while they increased their …nancing of investment spending using long-term debt. Thus, we can conclude that employing an index as a measure of treatment does not alter our results.

Addressing potential endogeneity concerns
This section considers an instrumental variable method (two-stage least squares 2SLS) to deal with the potential endogeneity of our explanatory variables and the bond market policy initiative. The identi…cation of the policy initiative requires an exogenous variable which is correlated with the policy of bond market development but does not directly impact …rms'access to external …nance. As plausible exogenous instruments for the policy initiative, legal origin of a country as British, French and German origin are implemented. Legal origin has also been used as an instrument for …nancial development by Liberti and Mian (2010) 18 . La porta et al. (2008) show that a country's legal origins based on British, French, German, or Scandinavian legal origins have a statistically large impact on its level of …nancial development. Beck et al. (2003) further stress that legal traditions of a country a¤ect the ability of a system to adjust to changing commercial requirements and encourages …nancial development of a country.
In addition to the policy initiative, it is also assumed that all control variables used in the model are endogenous and they are instrumented using their own values lagged twice. The validity and importance of the instruments for both the policy and other control variables are veri…ed using a number a diagnostic tests. The results for these 17 The index history statistics starts on 31/12/2000 and covers a variety of markets with small (Hong Kong, Singapore) and large (Korea, China) bond markets. Using simple weights will skew the index in favour of larger markets and reduce the weight of smaller markets. Hence, the baseline weight of these indices is adjusted by the local bond market size, sovereign local debt rating and GEMLOC investability indicator. 18 The data for legal origin are taken from La porta et al. (2008). tests are reported at the foot of the tables 19 . Table 8 shows the results of the 2SLS model. The results validate a signi…cant and positive impact of the policy initiative on …rms'access to long-term debt ratio and a negative impact on access to short-term debt ratio, with a stronger e¤ect on unconstrained …rms. Further, the results show that with an increase in …rms' access to long-term debt, their post-policy investment spending also increased, while post-policy investment declined for …rms dependent on short-term debt. Finally, the link between leverage and post-policy investment is much stronger for unconstrained …rms compared to their constrained counterparts. Other control variables maintain their signi…cance and expected signs.
Overall, the diagnostic tests given in Table 9 do not specify any problems regarding the application of instruments used and provide a reliable robustness check to our main results.

Additional control variables
While we have considered a wide set of explanatory variables in the main models, in this section we include additional control variables to ensure that our …ndings are not driven by omitted-variable bias. We include a dummy for the global …nancial crisis which takes value one for the period 2007-2010, and zero otherwise. We also control for other variables such as stock market capitalisation and global liquidity. The former is likely to be an important determinant of external …nancing as countries with larger stock markets help …rms to increase long term credit and access to external …nance (Demirgüç-Kunt and Maksimovic, 1999). The latter is measured by the cross-border credit growth in the Asia-Paci…c region. We include this variable to capture the market reactions to quantitative easing and tapering by United States on emerging economies in terms of capital ‡ows across borders. Thus, we try to disentangle the impact of the policy initiative on …rms' external …nancing by controlling for these additional variables.
The results are reported in Table 11 and con…rm that the policy did have a signi…cant impact on …rms'external …nancing. The results again con…rm that …rms increased their access to long-term debt and reduced their short-term debt. Further, …rms increase their investment spending using long-term debt while they reduce their investment spending using short-term debt. Finally, we …nd that both these relationships are stronger for unconstrained …rms compared to constrained …rms.

Excluding Korea
In order to con…rm that our main results are not driven by Korea, which has the largest bond market in our treated group, we remove it from our sample. The results are shown in Table 12 and we con…rm our main …ndings are both qualitatively and quantitatively very similar to our main results. Thus, we conclude that the inclusion of Korea in our sample does not bias our results in any way.

Alternative classi…cation of …rms
In our main empirical results, we classi…ed our …rms into constrained and unconstrained using the 50th percentile of the distribution in the pre-policy period. In order to con…rm that our results are not driven by the way we divide our sample, we use a robust framework of classi…cation schemes. Following Tsoukas (2011), we divide our …rms into constrained and unconstrained …rms using the 75th percentile as a cut-o¤ point in the pre-policy period. Thus, constrained …rms take value one if their pro…ts and coverage ratio are below the 75th percentile of the distribution of all the …rms in that particular year, and zero otherwise. Table 13 con…rm that the policy helped unconstrained …rms to expand their access to long-term debt and reduce their short-term debt much more compared to the …nancially constrained …rms. In addition, unconstrained …rms increased their investment spending using long-term debt much more compared to …rms which …nancially constrained. While in terms of short-term debt, there is no signi…cant di¤erence between constrained and unconstrained …rms with respect to their post-policy investment spending. Hence, we conclude that our results are robust to an alternative classi…cation of …rms.

Conclusion
After the Asian …nancial crisis of 1997-98, the need for developing local …nancial markets was realised in order to reduce over-dependence on a bank-dominated …nancial system and under-developed bond markets. Thus, in order to develop sound and more liquid bond markets, to prevent further capital account crises and the problem of 'original sin', Asian bond market initiatives were introduced in Asian economies 21 . Using a novel 21 The term 'original sin' was introduced by Eichengreen and Hausmann (1999) which means the inability of countries to borrow from abroad in their local currencies. It is a key factor of …nancial instability and possibility of default in a country. panel dataset on eight Asian countries over a period of 1996 to 2012, we analyse the impact of the Asian bond market initiatives on …rms' access to external …nance. The results based on the di¤erence-in-di¤erences method suggest that …rms'increased their access to long-term debt and reduced their short-term debts after the introduction of the ABF-2.
With respect to the …rm-level heterogeneity, the results show that the policy initiatives helped unconstrained …rms to increase their corporate bond issuances and reduce their bank …nance much more compared to their …nancially constrained counterparts. Next, we take into account the in ‡uence of …rms'external …nance on investment spending in the post-policy period. The results show that increased access to credit for …rms in the form of total long-term debt had a positive impact on …rms'investment spending. Finally, we …nd that with respect to long-term debt ratio, unconstrained …rms are able to increase their post-policy investment spending much more compared to constrained …rms due to their increased access to long-term debts after policy.
The results of this paper con…rm the fact that the Asian bond market initiatives helped in expanding the Asian domestic …nancial markets. Development of the …nancial bond markets helped …rms to achieve e¢ cient …nancing for business needs, encouraging investment and growth (Levinger et al., 2014). However, the level of development is widely diverse between di¤erent ASEAN countries. These policy initiatives have helped in expanding the local sovereign bond markets in Asia but the progress in terms of corporate bond markets is still low. In the past, policies have contributed to the development of domestic sovereign bond markets but very few have focused on the corporate bonds.
Thus, more progress is required for increasing diversi…ed issuer base so that …rms can receive funding from various sources without increasing shock volatility. When signi…cantly advanced, corporate bond markets can also have a huge e¤ect on investment and regional growth by supplying long-term funding.  Notes : The table presents sample   Notes : In column 1 the dependent variable is the short-term debt to total debt, while in column 2 the dependent variable is the long-term debt to total assets. Country dummies and time dummies interacted with industry dummies are included in the models with …xed e¤ects and standard errors clustered over …rms. Robust t-statistics are reported in the parentheses. Statistical signi…cance is denoted at 1% (***), 5% (**) and 10% (*). Notes : The p-value refers to the test of equality between constrained and unconstrained …rms. Robust t-statistics are reported in the parenthesis. Statistical signi…cance is denoted at 1% (***), 5% (**) and 10% (*). Also, see notes to Table   2. Notes : The dependent variable is …rm-level investment spending measured as the ratio of capital expenditures to total assets. 'Lev' is measured as short-term debt to total debt in column 1 and long-term debt to total assets in column 2.