Market-based fiscal discipline under evolving decentralisation: the case of Russian regions (1).

Timofeev, Andrey

INTRODUCTION

There are well-established arguments, both pro and con, on the matter of subnational borrowing. Debt financing has the potential to achieve intergenerational fairness by spreading the costs of building local infra structure over the entire period during which benefits from this investment occur. At the same time, subnational borrowing can be abused to shift the costs of local public services to future generations or other jurisdictions in excess of what would be justified by possible spillover of benefits. The common solution to this dilemma is to allow subnational borrowing subject to certain safeguards such as borrowing limits and explicit hierarchical authorisation for issuing debt. In countries where subnational governments are free from legislative or regulatory borrowing constraints by higher-level governments, they arguably face borrowing constraints imposed by lenders (Lane, 1993).

Capital markets can discipline governments by charging a higher risk premium as the volume of outstanding debt (explicit borrowing and payables) becomes too large relative to the expected revenue stream (Capeci, 1994). At some level of indebtedness the borrower will not be offered further credit at any interest rate. The economic literature identifies a number of general conditions for market-based fiscal discipline to work, the most critical of which is no expectations of a bailout. The latter is in part determined by intergovernmental arrangements that delineate competences and taxing powers between the levels of government.

Market-based discipline argument takes on new twists in transitional countries. On the one hand, at the early stages of transition a naive faith in the laissez-faire approach favoured market-based fiscal discipline in those countries. This trend was reinforced with the political drive for subnational autonomy, which among other things included borrowing powers. Therefore, many transitional countries have allowed subnational borrowing under very liberal regulations. On the other hand, the very conditions that are crucial for market-based fiscal discipline to work are often lacking in those countries. Thus, financial markets are often underdeveloped and lack liquidity. In addition, the institutions of local governance are immature and thus might not be conducive to fiscal policies responsive to rising borrowing costs.

For developed economies, empirical studies have documented market-based discipline in the form of positive relationship between the level of governments' indebtedness and their cost of borrowing (Bayoumi et al., 1995; Poterba and Rueben, 2001, for US state bonds; Capeci, 1994, for New Jersey municipal bonds). Although, the effectiveness of market-based fiscal discipline is particularly relevant for transitional economies as explained above, so far it has not been studied in the context of transition. Our study is the first to examine the performance of this market mechanism under the evolving institutions of decentralised governance in a transitional country. There are conjectures that market-based fiscal discipline might fail in the context of transition similar to the remarkable failure of coupon privatisation under the poor institutions of corporate governance in transitional economies.

The major challenge to an empirical test of market-based discipline is obtaining a consistent measure of market yields on the credit obligations of different jurisdictions. We have been able to identify one debt instrument that was utilised by a majority of Russian regions under standard conditions. These securities, called agrobonds, were issued by 69 out of a total of 89 regional governments in 1997-1998 to convert their outstanding liabilities to the federal government, who eventually auctioned off these bonds to private investors. We argue that the agrobonds' yield spread over the federal bond yield reflects the creditors' assessment of the risk that the issuer would not be able (or willing) to honor the obligation of servicing its agrobonds.

Over the last decade Russia has shifted from a permissive stance on subnational borrowing, which solely relied on market-based discipline, to administrative and rule-based controls on subnational borrowing (Martinez-Vazquez et al., 2006). This study attempts to shed some light on whether tightening of borrowing constraints had been warranted by incomplete decentralisation or whether it only reflects a change of political fashion.

The rest of this paper is organised as follows. In the next section, we examine the variation in intergovernmental arrangements faced by different regions in Russia. In particular we attempt to identify characteristics of a region that determine availability of a central government bailout. Section 3 reports the empirical evidence on the impact of these characteristics on the formation of risk premia by private lenders. Our conclusions follow.

INTERGOVERNMENTAL FISCAL ARRANGEMENTS IN RUSSIA

Similar to other countries, for the markets to effectively discipline governments in Russia a number of general conditions must be in place: free and open markets for credit; availability of information on the borrowers' accounts; the borrower's ability to promptly respond to market signals; and no expectations of a bailout. The latter condition 'appears to be the Achilles' heel of market discipline' according to policy studies (Lane, 1993). In addition to weakening the responsiveness of the borrower to market signals, the low credibility of the no-bailout policy also makes lenders discount the expected losses, which results in a failure to charge an adequate risk premium.

Russia has had some problems with each of the aforementioned conditions for market-based discipline, especially the credibility of the no-bailout policy (Martinez-Vazquez et al., 2006). However, the sensitivity of the federal government to local fiscal woes is limited to specific political concerns, such as wage arrears or disruption of vital services (eg central heating). The informal influence of subnational governments over the court system and weak enforcement of court judgments effectively means that regional government defaults can hardly lead to the disruption of vital services. At the same time the fact that in the recent past the federal government itself defaulted on its domestic bonds reveals its low concern over reputational or financial spillovers, which often prompt a central government bailout in other countries. Thus, ex ante the prospects of a federal bailout are uncertain and to a large extent depend on certain characteristics of a particular region, which we attempt to identify in this section.

The economic literature suggests that the credibility of the no-bailout policy is determined by a combination of factors related to either the central government's pay-off from bailing out a locality or the local government's costs of inflicting a fiscal crisis on itself (Inman, 2003). In particular, policy studies suggest that political costs of a local fiscal crisis (both to the local and central governments) depend on the relative responsibility of each level of government stemming from the division of authority in the system of intergovernmental fiscal relations (Rodden and Eskeland, 2003). Thus, sustainability of the no-bailout condition is in part determined by intergovernmental arrangements that delineate competencies and taxing powers between the levels of government.

For a bailout to occur, political arrangements should be such that at some point the central government cannot resist the pressure to help out a locality while the local officials can shift the blame to somebody else. This can happen when the central government shares responsibility for the public goods provided at the local level because it provides the bulk of local government revenue or when local governments do not have enough autonomy to undertake fiscal adjustments in response to a fiscal crisis. Voters might hold the central government responsible, because it has the means to resolve the crisis, unlike the local government, which may have never be given sufficient tax autonomy or which may have diverted or wasted all resources by the time of the crisis.

The Russian experience of fiscal reforms presents a typical example of incomplete and evolving decentralisation. At the start of transition Russia inherited a huge public sector, responsible for the provision of major social services including housing, transportation, healthcare, and education. (2) Trying to balance its fiscal accounts, the central government shifted the most onerous expenditure responsibilities down to regional governments (Bahl and Wallich, 1995, p. 347). (3) As subnational governments had no direct access to the central bank, and thus could not monetize their deficits, they were expected to accomplish the politically costly job of fiscal adjustment. This spontaneous decentralisation resulted in a lack of clearly defined competences of respective governments in the provision of affected services.

At the same time subnational governments have been given little revenue-raising authority. Although, regional and local governments may collect revenues from some taxes authorised by the federal government, they must fit their bases to the federal law, and may levy rates only within federal limits. In an average region, about 70% of the regional government revenues come from federal taxes, either through tax revenue retention at the point of collection or through redistribution via intergovernmental fiscal flows (see Table 1). Thus, on the margin, subnational governments have little capacity to mobilise revenues in response to a fiscal crisis. Therefore, as hypothesised by von Hagen and Eichengreen (1996), concentration of taxing authority at the central level can make the central government ultimately responsible for local fiscal outcomes.

This being said, we nevertheless should not expect all regional governments to engage in irresponsible borrowing and eventually be bailed out by the federal authorities. Indeed, the coefficients of variation presented in Table 1 reveal significant differences among regions in the extent of revenue autonomy. The category of 'own-source' revenue encompasses all revenue sources whose yield can be affected at the margin by regional governments, using their discretion to determine taxable bases or rates, or discretion to introduce the tax, or any combination of these three. By exercising this form of fiscal discretion, regional governments can respond to changes in the costs of service delivery and economic fluctuations. (4)

Different forms of revenue sharing also have different implications for the creditworthiness of regional governments. The category of 'assigned sources' of revenue refers to legislated long-term entitlements to (a share of) the regional yield from tax instruments over which regional officials have no discretion. (5) The category of 'regulated revenue' has historically referred to tax revenue sharing determined by the higher-level government as part of its annual budget process. However, since 1994, tax-sharing rates have been de facto standardised across regions and thus the regulated tax revenue effectively became akin to assigned revenue. The only discretion that subnational governments have with respect to the assigned revenue sources is deferring tax payments or offsetting tax liabilities with government payables. Although assigned revenue cannot be affected on the margin by regional governments, it possesses intertemporal predictability. This feature of assigned revenues helps regional governments to better budget and plan for future obligations and projects.

The final category, grants, is comprised of all intergovernmental transfers to regional governments whether formula-based or completely discretionary. Of course, these sources of revenue bring the least revenue autonomy to regional governments. (6) Figure 1 shows the growing importance of non-discretionary grants (equalisation grants, subventions, subsidies) in the total composition of federal transfers to regional governments. As can be seen from Figure 1, the share of non-discretionary grants in total federal transfers increased from 60% in 1995 to over 70% in 1999. (7) The remaining share is mostly accounted for by 'mutual settlements', which are typically not budgeted ex ante, but rather result ex post from emergency situations and political lobbying (Martinez-Vazquez and Boex, 2001). The relative decline of mutual settlements over time is an indication of improvements in the objectivity, stability and predictability of the federal-regional transfer system.

[FIGURE 1 OMITTED]

In summary, the Russian system of intergovernmental relations affects the credibility of the no-bailout policy. The unclear division of responsibilities between the levels of government, both in the legislation and voters' minds, makes the central government partially responsible for local fiscal outcomes. Furthermore, the centralisation of taxing authority at the federal level weakens the credibility of the central government's no-bailout commitment because the central government is both sensitive to local fiscal woes and possesses the means for rescuing troubled jurisdictions. Moreover, different regions seem to face different prospects of a federal bailout due to a wide variation in their characteristics that determine the availability of a federal bailout.

Thus, there is a wide variation among regions in their reliance on 'own-source' revenue and the incidence of ad hoc transfers from the federal government. It can be argued that regions with a smaller yield of their own sources of revenue are more likely to be bailed out because they lack flexibility for fiscal adjustment. In addition, regions receiving larger ad hoc transfers are more likely to be bailed out because the fiscal outcomes in these regions are presumably of higher concern for the federal government.

TESTING MARKET-BASED FISCAL DISCIPLINE HYPOTHESIS

The ultimate effectiveness of market-based fiscal discipline in Russia would manifest itself in sustainable volumes of debt accumulated by regional governments. This seems to be the case in most of Russian regions according to estimates reported in various studies (for a survey, see Martinez-Vazquez et al., 2006). The official figures, which became available only after the Budget Code of 2000 had required subnational governments to maintain debt ledgers, indicate that the total amount of explicit subnational debt has been below 5% of GDP. This is around one-third of the annual pre-transfer revenues of regional and local governments. Implicit debt, such as government guarantees and overdue payables, accounted for additional 2%-3% of GDP (this should also include extra-budgetary borrowing guaranteed by the government). (8) Thus, by all estimates, Russia's subnational debt falls below the levels observed in many other countries. (9)

While the average subnational debt has been rather low, several of Russia's regions have accumulated a stock of debt significantly exceeding the amount of their annual revenues. An extreme example is in the Republic of Tyva, where the stock of explicit debt exceeds pre-transfer revenue by a factor of five. Given that the maturity of debt that can be accommodated by the Russian credit markets is rather short, such debt-to-revenue ratios are hardly sustainable. For policy applications, it would be interesting to determine which component of the market mechanism failed in those regions. That is, it would be useful to see whether the markets failed to send corrective signals to the governments, or the regional governments failed to respond to these signals. Furthermore, it would be of practical interest to determine whether the cause of this failure lies in the design of intergovernmental arrangements.

Empirical Strategy and Testable Hypotheses

Our goal is to determine whether intergovernmental arrangements indeed interfere with the fiscal discipline imposed on regional governments by credit markets in Russia. Following previous studies, the risk premium on regional debt may be approximated with the following linear function: (10)

[[pi].sub.j] = [[alpha].sub.0] + [[alpha].sub.1][[beta].sub.j] + [[alpha].sub.2][[sigma].sub.j] + [X'.sub.j][beta] + [u.sub.j] (1)

where j indexes jurisdictions; b stands for the volume of outstanding debt normalised by the amount of recurrent revenues; [sigma] is the intergovernmental parameter signaling the probability of the central government bailout in the case of local government insolvency; [X.sub.i] is a vector of factors positively affecting the ratio of the fiscal surplus to the amount of recurrent revenues; and [u.sub.i] is the error term. The predictions are that [[alpha].sub.1] > 0, while [[alpha].sub.2], [[beta].sub.k] < 0. That is, higher levels of bailout signal [sigma] observed for the given jurisdiction at the time of lending result in lower risk premia charged by lenders.

Measuring the Risk Premia

The major challenge to an empirical test of market-based fiscal discipline is obtaining a consistent measure of market yields on the credit obligations of different jurisdictions. Although the bulk of regional debt in Russia is in the form of commercial lending, it is dominated by bank loans. Unlike for publicly traded securities, information on interest rates charged by banks is not publicly available. Fortunately, there is one debt instrument that was utilised by a majority of Russian regions under standard conditions (see the Annex for details). These securities, called agrobonds, were issued by 69 regional governments in 1997-1998 as a means to cover their guarantees on the commodity credits provided by the federal government to local agricultural producers in 1996. As Russian farmers conceived government credits as another form of subsidy, not many of them intended to repay the credits and thus regional government guarantees became due. Nine regions opted for paying off the guaranteed credit upfront, but the majority converted it into securities hoping that the federal government would not press too hard for repayment. However, the federal government auctioned off these bonds on the private market, which had had a 4-year experience with trading federal bonds (GKOs) totaling USD 32 billion.

Each region made three agrobond issues of equal size: of 1-, 2-, and 3-year maturity. All bonds had the same nominal value (RUR 10 thousand) and an annual coupon of 10%. On 20 June 1997, the federal government started offering agrobonds of different regions along with the minimum prices during special auction sessions at the Moscow Interbank Currency Exchange. (11) As can be seen from Figure 2, at the beginning the minimum prices set by the federal government for agrobonds of different regions were very close so that the associated yield ceilings clustered around the value of 6% points over the federal bonds' yield. (12) As a result, for the most creditworthy regions the entire issue was sold during one trading day while for others no bids exceeded the initial minimum prices. Figure 2 reveals how the federal government eventually had to lower the minimum prices (and thus raise yield ceilings) trying to sell some of the least demanded agrobonds. The federal government followed the same pricing strategy for most of the regions--daily incrementing the yield ceiling until it hit the market valuation. (13) It took on average 12 trading days for the yield ceiling to increase from the level of 6% points to 10% points over the GKO yield.

[FIGURE 2 OMITTED]

At the end, agrobonds of 26 regions were completely sold out and those of additional 31 regions were partially sold at the auctions. Out of almost RUR 7 billion (USD 1.2 billion) in agrobonds, roughly one-third of the agrobonds ended up held by foreigners, with another one-third held by Russia's largest banks and investment firms, while the federal government held the remaining unsold agrobonds (S&P, 2003). Thus, for 12 regions, agrobonds were not demanded under the imposed yield ceilings (going as high as 18% points over the federal bond yield).

Our strategy is to use the information on the discount rates at which agrobonds of a particular region were sold in auctions to approximate the default premium that this regional government would face in the credit market at that time. (14) We argue that the yield spread over the federal bond yield reflects the lenders' assessment of the risk that the issuer would not be able (or willing) to honor the obligation of servicing its agrobonds. Thus, our dependent variable is the agrobond yield spread averaged across all trading days in proportion to the fraction of this agrobond issue sold on each day. For the issues that did not register a single trade we only know that the market yield was above the ceiling imposed by the federal government. For such regions we set the dependent variable equal to the maximum yield ceiling that creditors were offered. (15) Thus, our dependent variable is essentially censored from above. Moreover, the censoring limits vary by region. This is because auction purchases stopped in the fall of 1997 as the world financial crisis hit the Russian market. This means that for those regions that issued their agrobonds late in August 1997, the federal government did not have enough time to complete the gradual incrementing of the yield ceilings to the clearing levels. (16)

We perform our empirical analysis on the sample of 1-year agrobonds--the largest sample (52 regions including 19 limit observation) out of the three maturities in our data set. (17) We have to acknowledge a sample selection problem due to several factors. First of all, 10 regions did not have any liabilities on the federal credits to be converted into agrobonds (eg cities of Moscow and St Petersburg). Second, nine regions opted for paying off the guaranteed credit upfront rather than converting it into securities. Finally, 16 regions issued their agrobonds after September 1997 and thus are not covered by our data set of the auction results. These 16 cases include both wealthy regions such as Nizhny Novgord and poor ones such as Dagestan. Thus, for eight of these 16 regions agrobonds were completely sold out by May 1998, while for the other eight regions the leftovers remained in federal ownership. Overall, we cannot see any pattern in the selection of our sample, and do not expect the sample selection to bias our results one way or another.

The auction results imply that on average investors felt that a Tambov Region agrobond should have a market yield only 5.05% points above the federal bond yield, while a Rhakasia Republic agrobond should bear a yield 14.55% points higher than the federal bond. For those regions whose agrobonds were at least partially sold at the auctions, the mean spread over the federal bond yield was 8.39% points, with a standard deviation of 2.35% points.

Measures of Indebtedness

The other major data requirement for this study is a measure of regions' indebtedness. The first official figures on the volume of outstanding explicit debt appeared in 2000 regional budget reports prepared in accord with the new Budget Code. The reported debt stock is broken down into commercial loans and intergovernmental loans. Before year 2000, budgetary reports had quoted only the annual flow of deficit financing broken down into detailed subcategories. Therefore, we construct the January 1997 stock of outstanding debt by subtracting the flows of deficit financing accumulated in the past from the recent figures on outstanding debt stock. (18)

To derive the measure of the relative size of debt, the nominal debt numbers are divided by the amount of a region's recurrent revenues (that is, total revenue excluding ad hoc grants). In our sample, the average relative debt is about 133%, half of which is accounted for by commercial lending. The standard deviation is 106%. The coefficient of correlation between commercial debt and intergovernmental loans is negative (-0.25). The Republic of Sakha, with the largest explicit debt, had a market yield of 9.17% points above the federal bond yield, which is 1% point higher than the average for the regions whose agrobonds were (partially) sold at the auctions. By contrast, Tambov Region, whose agrobonds were sold at the smallest yield spread, had relative debt of slightly above 50%, which is almost half of the average indebtedness. On the other hand, Khakasia Republic, whose agrobonds were sold at the largest yield spread, had relative debt around 36%of its recurrent revenues--only a quarter of the average indebtedness. Thus, the relation between indebtedness and interest rates is not straightforward and in fact can be part of multivariate relations. Nevertheless, we can note that the average indebtedness of those regions whose agrobonds were not demanded at the auctions was 30% points higher than the average indebtedness of other regions.

In addition to explicit borrowings, regional government liabilities also include overdue payables to employees and suppliers. For 1997, we have consolidated regional-local figures on overdue payables, which are broken down into salaries and payroll charges, transfers to population, and utility bills. (19) Relative to the regional government recurrent revenues, the total amount of regional-local budget arrears is not high--about 31%. This ratio is somewhat higher in regions whose agrobonds did not sell at the auctions--about 43%. Moreover budget arrears are weakly correlated with explicit debt: the coefficient of correlation is only 0.29.

Given the indirect way of measuring 1997 stock of outstanding debt, we also include an alternative measure of indebtedness: interest payments relative to the recurrent revenues in 1997. The coefficient of correlation between commercial debt and interest payments is 0.35. As a ratio to the regional government recurrent revenues, the average interest burden is about 1.26%, with the standard deviation of 2.17%. The average interest burden of those regions whose agrobonds did not sell at the auctions is only half of that for other regions. This could be indicative of either less borrowing or poorer payment discipline.

Bailout Signals

We use three complementary signals for the probability that the central government would provide a bailout if a regional government becomes insolvent. As pointed out by yon Hagen and Eichengreen (1996), the lack of revenue autonomy of subnational governments limits their ability to cope with a fiscal crisis on their own. These scholars argue that, given the inability of regional governments to undertake fiscal adjustment, the central government commitment to the no-bailout policy would not be rendered credible by lenders. Based on this argument, we use the share of 'own-source' revenue in the total revenue of the regional government as a signal for the probability that the central government would bail out this region. (20)

Our second signal for the availability of a bailout to the jurisdiction is its share in the national population. Wildasin (1997) predicts that availability of a bailout is negatively related to the fragmentation of local jurisdictions because externalities from discontinuation of local services are likely to be positively related to the locality's size. Moreover, his calculations provide at least one class of examples where there is a clear inverse relationship between jurisdiction size and availability of bailout. (21) At the same time larger localities are less likely to induce a debt crisis because they bear a larger share of the national government costs of a bailout (Goodspeed, 2002). In our sample, the average share of a region in the national population is 0.95 %, with a standard deviation of 0.76%. For those regions whose agrobonds did not sell at the auctions, the average population share is less than half of the other regions' average.

Finally, we use transfer dependence of regional governments as our third proxy for the availability of a bailout to the jurisdiction. Using Wildasin's (1997) framework, it can be shown that the amount of matching grants received by a jurisdiction reflects its spillover-generating capabilities and thus the central government's concern about the jurisdiction's fiscal outcome. We can also make an informal argument that, beyond matching grants, non-budgeted discretionary grants received by a region represent its political clout in the national capital. Even when the transfer size does not reflect the region's fiscal position but its bargaining power, the latter can be potentially utilised by the region to negotiate a bailout should a fiscal crisis occur. Thus, within Wildasin's framework, a larger transfer to a jurisdiction can signal to creditors a higher willingness of the central government to bailout a given jurisdiction.

However, in Russia not all grants are driven by the federal government's political concerns. As shown in Section 2, two-thirds of federal grants are accounted for by equalisation transfers, which are allocated based on the need/capacity formula. Indeed, the coefficient of correlation between the 'own-source' revenue yield and transfer dependence is 0.88. However, the share of discretionary (ad hoc) grants in the total amount of federal transfers received by a region is likely to better capture political favours of the federal government. Plainly, the discretionary component of transfers might in fact represent a bailout targeting distressed regions. The coefficient of correlation between the share of discretionary (ad hoc) grants and "own-source' revenue yield is 0.33.

For those regions whose agrobonds did not sell at the auctions, the average share of 'own-source' revenue in their total revenue was less than half of that observed in other regions. Also, in the regions with unsold agrobonds, the average share of discretionary grants in the total amount of federal transfers is only two-thirds of the share observed in the other regions. The 'own-source' revenue yield is negatively correlated with budget arrears on wages and social benefits (with a correlation coefficient of -0.3).

Econometric Issues

Given that the dependent variable is censored, OLS is an inappropriate technique for the estimation because it produces biased estimates. Compared to the most common alternative technique, MLE, OLS estimates are smaller in absolute value. (22) However, given that OLS coefficient estimates are not affected by other likely econometric issues, such as heteroskedasticity and non-normality, it can serve as a benchmark for evaluating estimates produced by alternative techniques. By contrast, for the Tobit model, a standard MLE technique to estimate equations with censored dependent variables, the consistency hinges on the assumption of normally distributed and homoskedastic errors. Moreover, some simulation evidence suggests that heteroskedasticity causes greater bias in maximum likelihood estimation than non-normality (eg Powell, 1986). To address this issue we estimate a Tobit model allowing multiplicative heteroskedasticity in the form of [[sigma].sup.2.sub.i] = [[sigma].sup.2]exp(2[gamma][z.sub.i]), where [z.sub.i] is the region i's share in the federal tax base.

By regressing regional governments' costs of borrowing on their levels of indebtedness along with other regional characteristics, we assess the effectiveness of the supply side of the market mechanism. That is, we check whether credit markets indeed send corrective signals to borrowers. However, there is a reverse causality from the costs of borrowing to government demand for credit. In non-censored models, in order to eliminate the endogeneity bias, we would have to instrument regional governments' demand for credit with some exogenous variables. That is, we would have to find variables that affect regional governments' costs of borrowing only through the default risk stemming from the level of indebtedness. Any discrepancy in estimates obtained with instrumented and non-instrumented regressions would indicate some responsiveness of subnational borrowing to rising interest rates.

Unfortunately, in limited dependent variables models, it is very difficult to deal with endogeneity. Unless strong assumptions are made on the exact relationship between the endogenous regressors and the instruments, it is generally not possible to apply instrumental variable techniques. Thus, our strategy is to include the endogenous level of debt among our explanatory variables hoping that the endogeneity would not bias coefficient estimates for variables other than the debt level.

Apart from the endogeneity, the estimation procedure is further complicated with censoring limits varying by region. Indeed, we observe risk premia resulting from the auction only if they fit the maximum allowed yield spread as determined by the minimum price set by the federal government for a particular agrobond issue on that particular day. Unfortunately, many censored models (such as semiparametric censored models) assume a uniform censoring threshold. (23) Therefore, our primary technique is Tobit with multiplicative heteroskedasticity, which allows for varying censoring limits.

Estimation Results

As we do not have many observations, we have to be parsimonious about the number of regressors included. Many of our variables are correlated between each other, which results in insignificant estimates of individual impacts when these variables are jointly included in the regression. Figure 3 shows the statistical relations among our variables in the form of a principal component biplot chart. (24) The biplot display is a commonly used multivariate method for graphing row and column elements (in this case, regions and their characteristics correspondingly) using a single display (Gabriel, 1971). The rays originating from the center of the graph are linear projections of our variables onto the two-dimensional space defined by the two principal components--two largest eigenvalues of our data set. Thus, most variability in the original multidimensional data set occurs in the chosen two-dimensional space. Variable rays representing uncorrelated variables are orthogonal. The smaller the inner angle between rays, the higher is the positive correlation between the values of the corresponding variables. For negatively correlated variables, the inner angle is greater than 90[degrees]. Longer rays represent variables with larger standard deviations.

[FIGURE 3 OMITTED]

The goodness of fit of our two-dimensional projection, defined as the fraction of the sum of squares of singular values accounted for by the two largest singular values, is 0.476. Using the above stated rule of thumb for interpreting a biplot chart, we can conclude that the share of 'own-source' revenue, per capita revenue relative to the subsistence level, the share of ad hoc grants, and population size are positively correlated. In addition, transfer dependence of regions is almost perfectly correlated to the share of 'own-source' revenue with a negative sign. Therefore, while including in the regression equation only those among the related variables whose coefficient estimates have the highest statistical significance, we should remember that these estimates also capture the effect of other related variables.

Thus, the share of 'own-source' revenue is not included in the final regression because of being statistically insignificant when included jointly with per capita revenues. Similarly, none of the different types of implicit liabilities are included in the presented regressions as they were not statistically significant in any of the employed specifications. Out of various control variables suggested in earlier studies (per capita revenue, unemployment, volume of the bond issue), our final regressions include only the per capita amount of pre-transfer revenues adjusted for the regional subsistence level.

The biplot chart also suggests that the yield spread has only little positive correlation with the stock of commercial debt and the burden of debt service. Also, the yield spread has only little negative correlation with intergovernmental loans and social arrears. Thus unless the statistical relations change in a multivariate setup, we can expect some of our proposed regressors to have little explanatory power for the yield spread. Note however a strong correlation between outstanding debt and outlays on debt service, which provides us some comfort regarding the indirect way of measuring debt. Later in this section we will use the biplot chart again to identify clusters of similar regions.

The second column of Table 2 reports estimation of Eq. (1) by means of OLS with White's standard errors. The impact of the commercial debt stock--measured relative to the annual recurrent revenue--is not statistically significant. Interest expenditures relative to the annual recurrent revenue are positively related to the yield spread and this relation is statistically significant at the 10% level. Adjusted per capita revenue has a negative coefficient, which is statistically significant at the 10 % level The other bailout signals--the share of the discretionary grants in the total grants received and the region's share in the national population--have no statistical impact on the yield spread.

The fourth column of Table 2 reports the results of estimating Eq. (1) using the homoskedastic Tobit model. This estimation produces 10%-significant coefficients only for population size and per capita revenue. (25) Both coefficients are negative, which is consistent with our arguments regarding the bailout availability in the first case and observed capacity to generate a fiscal surplus in the latter case. The results of the heteroskedastic Tobit estimation are reported in the last column of Table 2. In addition to the negative coefficients for the population size and per capita revenue, the heteroskedastic Tobit also produces 10%-significant coefficients for the commercial debt and discretionary grants. However, the signs of these coefficients--negative and positive correspondingly--are counter to our predictions.

For the purpose of comparability, we also re-estimated the linear regression under the assumption of multiplicative heteroskedasticity [[sigma].sup.2.sub.i] = [[sigma].sup.2]exp(2[gamma][z.sub.i]) and presented the results in the third column of Table 2. The results are roughly consistent with those from the heteroskedastic Tobit except that the positive coefficient of the interest expenditures becomes statistically significant at the 1% level. Note that in accordance with common wisdom, the Tobit model produces larger estimates in absolute value than the linear regression.

Overall, the econometric results do not provide a clear-cut answer regarding the causalities between the indebtedness and interest costs. The negative coefficient on the commercial debt can be due to the reverse causality from lower interest rates to larger borrowing. Similarly, while representing past borrowings, the current interest payments can be affected by reverse causality from the current default premia in the case of autocorrelation exhibited by the latter.

There is however more clarity in our results on the bailout signals, which is the focus of this study. According to our results, lenders do not see ad hoc grants as a reason to discount risks of a possible default. Instead, lenders seem to favour large and affluent regions. This general pattern can be also seen from the biplot chart (Figure 3). The individual points in the chart are linear projections of our observations labeled with corresponding region codes. As the variables are standardised by subtracting the mean and dividing by the standard deviation, data points located in the center of the graph represent regions with average values of the variables. Data points located away from the center in the direction of some variable ray represent regions with values of that variable that are distinct from the average.

We can identify two clusters of observations on the chart. The first, larger cluster represents regions with higher dependence on federal transfers, larger amounts of overdue utility bills and larger yield spreads on agrobonds. Note that the most troublesome region (Republic of Tyva, code 70) protrudes the furthest from the average along the yield spread ray on the biplot graph. The second, smaller cluster gathers regions with a larger share of own-source revenue, larger per capita revenue relative to the subsistence level, and also a larger share of discretionary grants in total federal transfers. These latter regions also happen to have larger population.

In summary, we seem to detect that larger and wealthier regions receive more grants in the ad hoc form and undertake more borrowing at smaller interest rates and as a result incur more interest expenditures relative to their recurrent revenue. Moreover, after we control for the size and revenue capacity, we detect a positive relation of the yield spread to interest expenditures and ad hoc grants. This pattern is compatible with some of our predictions but runs counter to others. The results call for further testing when better data become available.

CONCLUSIONS

In theory, credit markets can potentially correct irresponsible fiscal behaviour by charging adequate risk premia or excluding a profligate jurisdiction from further borrowing altogether. However, to be effective in a particular country, market forces require several general conditions to be present in its system of intergovernmental fiscal arrangements. This study evaluates the impact of intergovernmental arrangements on the performance of market-based fiscal discipline in Russia. To our knowledge, this is the first test of market-based fiscal discipline on data from a transitional country. Although some caution should be taken due to the limitations of the available data, certain conclusions can be drawn from the results.

The impact of intergovernmental factors on the formation of risk premia by the credit markets seems to go beyond the prospects of a bailout. In fact, a greater revenue sufficiency of regional governments appears to lower their costs of borrowing. This is consistent with other studies finding a positive relation of the borrowing costs to transfer dependence (Capeci, 1994) and a negative relation to the revenue base (Capeci, 1994) and primary surplus (Caselli et al., 1998). This is despite the fact that debt is measured relative to the revenue base and thus the latter should not matter for risk assessment. Thus, it appears that creditors favour wealthier jurisdictions notwithstanding their debt burden.

One can argue that larger revenue autonomy is interpreted by the lenders as the ability of the borrower to undertake a revenue-raising effort and generate fiscal surplus sufficient for the repayment of the debt. Thus, although relatively large revenue autonomy makes a federal bailout less likely for such a region, the lenders might still favour this borrower because he is less likely to become insolvent in the first place.

While the negative relation between a region's size and its borrowing costs detected in this study can be indicative of the bailout prospect ('too big to fail'), it can also be explained with the fact that larger regions tend to be wealthier and thus have better ability to pay. (26) This is also consistent with the observation by Petersen and Huertes (2003) stating that 'in the absence of information and experience, the 'name' and size of a subnational government have had disproportionate importance'. Overall, we can conclude that even if the intergovernmental factors had some effect on the formation of risk premia, it was too weak to override the low creditworthiness of poorer regions.

The major policy implication of this study is that incomplete decentralisation in Russia did not justify suppression of the market discipline with administrative controls. Even when intergovernmental arrangements suggested possibility of a bailout, lenders chose not to count on such a possibility and denied credit to those regions. It appears that rather than constraining market borrowing, the central government needs to commit itself to limit intergovernmental loans provided to those less creditworthy regions. Indeed, we find that commercial debt has almost perfect negative correlation with intergovernmental loans. Moreover, we uncover the strategy that allows regional governments to obtain intergovernmental loans. We see that intergovernmental loans are closely correlated with arrears on wages and social payments such as stipends.

This study also suggests that larger regions might require more stringent hierarchical supervision because lenders seem to discount default risks for such regions compared to smaller regions with a similar level of indebtedness.

Annex

Agrobonds

Agrobond securities were issued by 69 regional governments during 19971998 as a means to cover regional guarantees on commodity credits provided by the federal government to local agricultural producers in 1996. As Russian farmers viewed government credits as another form of subsidy, not many of them intended to repay the credits and thus regional government guarantees became due. Some regions opted for paying to the federal government upfront the credits that they had guaranteed, but the majority converted it into securities hoping that the federal government would not press too hard for payment. However, the federal government auctioned off these bonds to private investors, at a discount of about 20%. Out of almost RUR 7 billion in agrobonds, roughly one-third ended up held by foreigners, with another one-third held by Russia's largest banks and investment firms and the rest (unsold) held by the federal government.

Each region made three agrobond issues of equal size: 1-year, 2-year, and 3-year bonds. The bonds were not collateralised but unconditionally secured with all the assets of the borrower. All bonds had the same nominal value (RUR 10,000) and an annual (or two semiannual) coupon of 10%. It follows from this that all issues of agrobonds should have been redeemed by June 2001. However, by December 2001 only 42 regions had paid off the principal and the interest on the 1-year issue. Moreover, only 27 regions paid off the principal and both coupons of the 2-year issue. As for the 3-year issue, only 24 regions paid the principal and all three coupons. Twelve regions did not make a single payment on any of the agrobond issues.

In July 1998, agrobond holders complained to federal authorities about the regional governments' delay in payments. In response, the federal Ministry of Finance declined to intervene, citing the budget autonomy of regions. The only concession made by the federal government on the issue was easing the regions' debt burden by restructuring regional liabilities on the portion of agrobonds that remained in federal ownership.

With the exception of the Karachayevo-Circassian Republic, none of the 12 regions whose agrobonds failed to find buyers at the auctions and thus remained completely in federal ownership have made a single payment on these liabilities. The Karachayevo-Circassian Republic presents an exception as the liabilities of its agrobonds were offset against special federal budget appropriations for drought aid to the Republic in 1998.

Out of 26 regions whose securities were completely sold out at the auctions, only 18 had cleared them all by December 2001. Many regions whose agrobonds were partially sold in 1997 have cleared the securities held by private investors while refusing to service the portion held by the federal government. This can be explained in part by the fact that some law firms buy defaulted agrobonds for pennies on the secondary market and aggressively seek settlement through courts with varying degrees of success. In addition, regional governments' debtors often buy agrobonds at a significant discount to pay off their debt to regions at face value. It should be mentioned that the ruble devaluation of 1998 significantly reduced the real value of agrobonds.

REFERENCES

Bahl, RW and Wallich, CI. 1995: Intergovernmental fiscal relations in the Russian federation. In: Bird, RM, Ebel, RD and Wallich, CI. (eds). Decentralization of the Socialist State: Intergovernmental Finance in Transition Economies. Regional and Sectoral Studies. World Bank: Washington, DC.

Bayoumi, T, Goldstein, M and Woglom, G. 1995: Do credit markets discipline sovereign borrowers? Evidence from US states. Journal of Money, Credit and Banking 27: 1046-1059.

Capeci, J. 1994: Local fiscal policies, default risk, and municipal borrowing costs. Journal of Public Economics 53(1): 73-89.

Caselli, F, Giovannini, A and Lane, T. 1998: Fiscal discipline and the cost of public debt service: Some estimates for OECD countries. International Monetary Fund Working Paper WP/98/55.

Gabriel, KR. 1971: The biplot graphic display of matrices with application to principal component analysis. Biometrika 58: 453-467.

Goodspeed, TJ. 2002: Bailouts in a federation. International Tax and Public Finance 9: 409-421.

Greene, WH. 1999: Econometric Analysis. 4th Edition. Prentice-Hall: New York.

Inman, RP. 2003: Transfers and bailouts: Enforcing local fiscal discipline with lessons from US federalism. In: Rodden, J, Eskeland, G and Litvack, J. (eds). Fiscal Decentralization and the Challenge of Hard Budget Constraints. MIT Press: Cambridge and London.

Lane, TD. 1993: Market discipline. International Monetary Fund Staff Papers 40(1): 53-88.

Lipkovich, I and Smith, EP. 2002: Biplot and singular value decomposition Macros for Excel[C]. Journal of Statistical Software 7: 5.

Martinez-Vazquez, J and Boex, JLF. 2001: Russia's Transition to a New Federalism. WBI Learning Resources Series. The World Bank: Washington, DC.

Martinez-Vazquez, J, Timofeev, A and Boex, JLF. 2006: Reforming Regional-Local Finance in Russia. WBI Learning Resources Series, The World Bank: Washington, DC.

Myazina, E. 2005: Renewed interest in regions. Vedomosti. 29 April 2005.

OECD. 1999: Revenue Statistics: 1965-1998: Special Features, Taxing Powers of State and Local Government, the Interpretation of Tax-to-GDP Ratios, the Impact of GDP Revisions on Reported Tax Levels. Organization for Economic Co-operation and Development: Paris.

Petersen, J and Huertes, M. 2003: Designing and implementing credit assistance to subnational governments. In: Mila, F, Petersen, J. (eds). (2004) Sabnational Capital Markets in Developing Countries: From Theory to Practice. World Bank, Oxford University Press: Washington, DC, New York and Oxford.

Poterba, JM and Rueben, KS. 2001: Fiscal news, state budget rules, and tax-exempt bond yields. Journal of Urban Economics 50(3): 537.

Powell, JL. 1986: Symmetrically Trimmed Least Squares Estimation of Tobit Models. Econometrica 54: 1435-1460.

Rodden, J and Eskeland, G. 2003: Lessons and conclusions. In: Rodden, J, Eskeland, G and Litvack, J. (eds). Fiscal Decentralizatinn and the Challenge of Hard Budget Constraints. MIT Press: Cambridge and London.

Standard & Poor's. 2003: Surviving after default: The experiences of local governments in Argentina and Russia, 25 February 2003.

von Hagen, J and Eichengreen, B. 1996: Federalism, fiscal restraints, and European Monetary Union. American Economic Review 86(2): 134-138.

Wildasin, DE. 1997: Externalities and bailouts: Hard and soft budget constraints in intergoverrunental fiscal relations. Policy Research Working Paper WPS1843. The World Bank: Washington, DC.

(1) I am grateful to Olga Shirokova, Econometric Unit Vedi (Moscow) for supplying some of the information used in this study. I am also grateful to Roland Anderson, Andrew Austin, Stepan Jurajda, Jorge Martinez-Vazquez, Saloua Sehili, Jan Svejnar, and participants of Ronald Coase Institute's Workshop in Institutional Analysis, 2002, and the 61st Congress of the International Institute of Public Finance, 2005, for helpful comments and discussions. All remaining errors are solely mine.

(2) In 1992, consolidated budget expenditures accounted for about 30% of GDP. In fact, the extent of public provision was even larger, in the Soviet system many basic goods and services were provided by state-owned enterprises as fringe benefits to their employees. Hence, before being privatised such enterprises maintained huge social assets: housing, kindergartens, hospitals, and recreation facilities. Privatisation was accompanied by the process of divestiture, meaning a transfer of social assets and the responsibility for their financing to municipalities.

(3) In turn, regional governments had complete discretion to download these responsibilities further to the local level.

(4) With an average share of 'own-source' revenue equal to 21% of total revenue in 1997, Russian regions had considerably less revenue autonomy than their counterparts in the US, Switzerland, and Belgium, but nevertheless more than in Spain, Germany, and Mexico (OECD, 1999).

(5) These statements would need to be qualified if the local authorities can affect the behaviour and diligence of federal tax agents, who are responsible for collecting revenues to all levels of government. Although tax policy and tax administration are highly centralised according to the formal system, local authorities could use their informal influence on federal tax agents to affect the rigour of tax enforcement towards local enterprises.

(6) Strictly speaking not all grants should be considered as equally negative in terms of revenue autonomy. For example, an equalisation grant based on stable formulas for both funding and distribution of the funds can bring much more stability and predictability to local budgets than ad hoc specific grants allocated every year.

(7) Until 1998, the 'subventions' category reported compensation to the City of Moscow for the costs related to its status of a national capital. In 1999, the 'subventions' category accounted for the federal governments' aid to regions in stocking up the supplies of necessities in localities of the Far North (being a relic of the Soviet planners' decision on location, northern settlements have to be subsidised on humanitarian grounds before eventual retreat is carried out). The 1999 figures do not sum up to 100% due to the presence of unidentified 'other' transfers.

(8) Although budget execution at the subnational level is still reported on a cash basis in Russia, by 1997 many subnational governments had started reporting accumulated payables to the higher-level government in a separate file according to the same classification used for actual outlays. With the adoption of the new Budget Code all subnational governments have been required to maintain a ledger of purchases registering terms of order and information about the supplier.

(9) This statement holds even after accounting for the fact that the ruble devaluation of 1998 significantly reduced the real value of subnational debt.

(10) Our specification is similar to Capeci (1994). We do not use Bayoumi et al.'s (1995) nonlinear specification, which was derived for one functional class of probability distribution. We did however attempt to include non-linear terms in our equation but did not find any improvements in estimation.

(11) Data on agrobond trades were provided by the Econometric Unit Vedi (Moscow).

(12) For the benchmark federal bond (GKO076), the yield was fluctuating between 18.2% and 20.3% during June-September 1997.

(13) The auction was structured in terms of the agrobond price. However, our analysis (and presumably investors' decision) is based on the associated yield spread that is the difference between the yield to the maturity on the agrobond and the yield on a federal bond of a similar maturity taken on the day of the auction.

(14) While agrobonds give us a unique opportunity to study yield spreads on a large sample of regions, it also limits our study to the pre-financial crisis period. Although by 2005 the subnational bond market completely revived in terms of trading volume, it has never represented such a broad number of regions. As of 2005, only about 35 subnational governments had its bonds traded on the market (Myazina, 2005). Although the looming financial crisis affected the ability of central government to bailout regions, the cross-regional nature of our analysis allows us to control for such common factors while assessing the impact of cross-regional differences in intergovernmental arrangements.

(15) Lower yield ceilings set for prior trading sessions for which no purchases took place do not convey any additional information.

(16) For seven out of the 19 unsold 1-year agrobonds the final ceilings were around 13% points over the GKO yield. For the remaining 12 issues the final yield ceilings seem to be evenly spread from 9% to 18% points over the GKO yield. At the same time, for one completely sold issue the final yield ceilings exceeded 13% points premium: Republic of Khakasia--14.55% over the GKO yield.

(17) Pulling together observations from the three different maturities would have added only one region to the sample, while introducing considerable difficulties in comparing risk premia across maturities.

(18) As before year 2000 regional governments were not required to include outstanding debt in their budgetary reports, the relation between indebtedness and risk premia is a product of creditors' ability to both obtain this information from other sources and utilise it for risk assessment.

(19) Post-2000 data indicate that regional governments account for less than 20% of subnational budget arrears (Martinez-Vazquez et al., 2006). Nevertheless, consolidated regional-local budget arrears can serve as a proxy for regional government liabilities to the extent that they are used by local governments as a strategic tool for extracting regional assistance. The effectiveness of local governments' arrears as a tool to squeeze regional funds ranges from quite high for wages and social payments to rather low for utility bills.

(20) By 'own-source' we mean all revenue sources whose yield can be affected at the margin by regional governments, using their discretion to determine taxable bases or rates, or discretion to introduce the tax, or any combination of these three.

(21) Wildasin's (1997) calculations provide at least one class of examples where there is a clear inverse relationship between jurisdictional

size and availability of bailout. However, this is true in a one-shot game. In a sequential interaction, the central government might choose to deny a bailout to a large locality in order to send a strong signal to other jurisdictions and thus build a reputation for fiscal discipline.

(22) An empirical regularity is that the MLEs can be approximated by dividing the OLS estimates by the proportion of non-censored observations in the sample (Greene, p. 697).

(23) Theoretically, we can subtract these varying censoring limits from both sides of the regression equation and thus arrive to the uniform (zero) censoring limit. However, because the yield ceilings gradually approach the clearing levels, they are likely to be correlated with the same variables that determine the market risk premia (ie, indebtedness and revenue autonomy). Thus, if the censoring limits were explicitly accounted for on the RHS of our equation, we would run into the multi-colinearity problem.

(24) The biplot display is drawn using Excel Macros from Lipkovich and Smith (2002).

(25) We are interested in the true market yield spread, which is the latent variable in the Tobit model. Thus, the marginal effects of our interest are the estimated Tobit coefficients.

(26) Creditors might favour larger jurisdictions because of more sophisticated financial management and economies of scale in fixed costs of debt transactions.

ANDREY TIMOFEEV

Andrew Young School of Policy Studies, Georgia State University, PO Box 3992, Atlanta, GA 30302-3992, USA. E-mail: atimofeev@gsu.edu

Table 1: Average structure of regional government revenue

 1997 1998 1999

Own-source revenue 21.26%# 23.6%# 31.0%#
 out of which: (0.70)# (0.67)# (0.60)#
Corporate income tax 11.22% 11.66% 18.12%
 (0.80) (0.84) (0.79)
Sales tax -- 0.16% 2.46%
 -- (2.89) (0.91)
Enterprise assets tax 8.54% 9.74% 7.08%
 (0.84) (0.80) (0.69)
Non-tax 0.99% 1.04% 1.80%
 (1.61) (1.02) (1.31)
Assigned revenue 3.66%# 3.26%# 3.99%#
 out of which: (2.19)# (2.48)# (2.40)#
Levies on subsoil users 2.65% 2.94% 3.64%
 (3.04) (2.76) (2.64)
Regulated revenue 20.76%# 24.01%# 20.67%#
 out of which: (0.58)# (0.55)# (0.58)#
VAT 9.61% NA 8.02%
 (0.78) NA (0.86)
Personal income tax 4.57% NA 6.20%
 (1.02) NA (0.74)
Excises 3.59% NA 5.35%
 (1.06) NA (0.99)
Grants 54.32%# 49.12%# 44.37%#
 (0.48)# (0.57)# (0.62)#

Note: The main subtotals indicated with #.

Notes: Coefficients of variation are provided in parentheses.

Source: Author's calculation based on data from the Ministry of
Finance.

Bold emphasizes the main subtotals.

Table 2: Estimation results, dependent variable: yield spread
(% points)

 OLS Multiplicative
 (robust heteroskedastic
 errors) regression

Constant 11.72 *** (10.06) 12.29 *** (0.85)
Commercial -0.65 (0.48) -1.07 *** (0.34)
 debt
Interest 21.66 * (13.02) 50.6 *** (9.15)
 expenditures
Per capita -5.74 * (4.40) -12.46 *** (2.18)
 revenue
Population -49.73 (46.61) -89.2 *** (28.5)
 share
Soft grants -0.83 (1.50) 2.59 *** (0.89)
[R.sup.2] 0.14
Heteroskedasticity -- -272.14 *** (42.18)
 term
[SIGMA] 0.047 *** (0.007)
# of observations 52 52
Out of which 19 19
 are censored

 Tobit Heteroskedastic
 tobit

Constant 14.31 *** (1.40) 14.51 *** (1.53)
Commercial -0.68 (0.85) -1.31 * (0.71)
 debt
Interest 18.78 (27.49) 51.20 (32.28)
 expenditures
Per capita -10.34 ** (4.98) -16.88 *** (5.97)
 revenue
Population -122.65 * (76.16) -153.8 ** (65.35)
 share
Soft grants -0.07 (2.82) 3.95 * (2.34)
[R.sup.2]
Heteroskedasticity -- -200.65 *** (41.12)
 term
[SIGMA] 0.037 *** (0.005) 0.092 *** (0.030)
# of observations 52 52
Out of which 19 19
 are censored

Notes: * statistically significant at the 10% level; ** statistically
significant at the 5% level; *** statistically significant at the 1%
level.