Fiscal Policy, Sovereign Debt and Default with Model Misspecification Viktor Tsyrennikov∗ Cornell University February 5, 2014 Abstract We study a small open economy model with production and public good provision. The government can borrow from abroad but it is subject to a premium reflecting probability of default. We show that when both creditors and the country-debtor have fears of model misspecification level and volatility of the sovereign bond premium increase significantly. This stems from the endogenous disagreement between the agents taking opposite sides of financial trade. Pessimism stemming from model misspecification leads the government to use taxes more actively. Out of precaution the government taxes more in good times and “manages the adverse states” by increasing taxes heavily. Default periods are associated with a significant increase of government’s savings as observed in the data. Keywords: sovereign default, debt, fiscal policy ∗ email: vt68@cornell.edu. I would like to thank ... 1 1 Introduction We study endogenous fiscal policy and default in a small open economy. The economy is populated by a representative household that consumes and supplies labor to the market. The household cannot borrow and all borrowing is done by the government. The government, in turn, chooses labor tax, public good provision, and pubic debt to maximize society’s life-time utility. Government budget does not have to be balanced every period and deficits (or surpluses) can be financed by borrowing from (or lent to) international creditors. But when the circumstances lead the government to accumulate significant debt it may choose to default. The creditors anticipate that the government may repudiate its debt and price in a possibility of sovereign default. Unlike Arellano [2008] and Aguiar and Gopinath [2006] we study a production economy with an endogenously determined fiscal policy. In section 2 we argue that government’s action’s, in particular deteriorating budget balance was one of major reasons for default in Argentina in 2001. Default in this model is driven by a government’s reluctance to increase taxes or decrease public spending. Government’s taxes are a great policy tool in normal times. But when productivity is low higher taxes add to default worries by reducing output while increasing tax revenue only marginally. Provision of public goods is reduced gradually at times of distress to maintain households’ welfare and not to add to fears of default. Both households in the country-borrower and creditors are assumed to have recursive utilities. We study a special case with robust (multiplier) preferences as in Hansen and Sargent [2007]. This preference specification locks in the intertemporal elasticity of substitution (IES) at 1 but allows varying the risk-aversion parameter that we denote by θb . Hansen and Sargent interpret θb as a measure of concern for model mis-specification. Consider the following thought experiment. An agent believes that distribution F is a reasonable description of data. But he realizes that there are other choices F̃ that could also fit the data reasonably well. An agent with rational expectations behaves as if F is the true model. An agent with concerns for model mis-specification behaves as if the true model is the worst possible assign- 2 ment among possible F̃ . The larger is θb the more different F and F̃ are. Distribution F̃ is a “pessimistic” version of F . That is it assigns more weight to adverse, from the agent’s point of view, outcomes. Creditors’ pessimism increases the country’s borrowing cost by a “pessimism” premium. We would like to highlight that preference for mis-specification is twosided. This leads to endogenous disagreement between the governmentborrower and creditors who are on the opposite sides of sovereign debt transactions. The endogenous disagreement increases volatility of bond returns. Disagreement is present even if only one of the agents has concerns for model mis-specification. Consider the case with pessimistic creditors and a rational borrower. Because of the payoff structure of the defaultable sovereign debt1 disagreement is only about the “far” left tail of the distribution. When lowest productivity states occur premia react dramatically. But these states occur only infrequently; so, volatility of premia increases only marginally. Next, consider the case with rational creditors and a pessimistic borrower. Bond prices are closer to beliefs and valuation of the wealthier market participant. So, as wealth moves between creditors and the borrower bond prices and returns change significantly. In a closely related work Presno and Pouzo (2013) extend the analysis in Arellano [2008] and assume that creditors, but not borrowers, have concerns about model mis-specification. Pessimistic creditors behave in many ways like impatient agents. To achieve a reasonable default probability households have to be made unrealistically impatient. With two-sided concern for model mis-specification the model performs well even with small differences in discount factors.2 Adam and Grill [2011] study default decisions in the presence of disaster events. Disaster events, like preference for robustness, generate caution on the part of both creditors and borrowers. It is another way of introducing (two-sided) pessimism into economic behavior. Aguiar and Amador [2011] also study a model with production and default. In their model the government, for political reasons, overvalues current 1 It pays only in high productivity states. When creditors are pessimistic the borrower has to be made very impatient so that he would default in equilibrium. When both creditors and the borrower are pessimistic the borrower would default even if he were relatively patient. 2 3 spending. While the government cannot commit to repaying its obligations (endogenously incomplete markets) default does not occur in equilibrium. In this way our model is more suitable for quantitative analysis. But it would be fruitful to incorporate the political economy frictions into the current model. Pouzo [2013] also studies optimal fiscal policy with production and default under incomplete markets. It builds upon (a closed economy of) Aiyagari et al. [2002] but relaxes the assumption that the government can commit to a tax and debt policy. The government may choose to repay its debt only partially. Default does arise in equilibrium but it is a levi on domestic households rather than foreign creditors. In what follows we present data facts about the 2001 default episode in Argentina to justify some of the modelling choices that we make. Description of the model follows. Then we present our numerical results and conclude. 2 Data facts Most of the research on sovereign crises concentrates on the three facts. First, output and consumption are very volatile and strongly correlated. The two series are plotted in figure 1 panel A. Second, trade balance and current account (as shares of GDP) are strongly counter-cyclical. These series are plotted in figure 4 panel B. Third, cost of borrowing for emerging markets is high and extremely volatile. While the first two facts can be rationalized within an exchange economy with incomplete markets and default as in Arellano [2008] and Aguiar and Gopinath [2006]. But these models are unsuccessful in matching the facts about country premia. This work aims to provide a rationale for the observed country premia. We start by pointing out that borrowing in emerging market economies is done primarily by a government not private agents. We illustrate this point using the data on Argentina. Figure 2 panel A shows outstanding public and private liabilities as percentages of Argentina’s GDP. Private liabilities are a small fraction, less than 20%, of the total liabilities. Panel B of the same figure shows net private borrowing from offshore banks. Apart from 4 A. Output and consumption B. Cost of borrowing - USD LIBOR 0.15 0.80 EMBI Argentina 0.70 0.10 0.60 0.05 0.50 0.00 0.40 0.30 -0.05 0.20 -0.10 output consumption 0.10 -0.15 0.00 1992 1994 1996 1998 2000 2002 2004 2006 1996 1998 2000 2002 2004 2006 Figure 1: Argentina’s output, consumption and cost of borrowing A. Outstanding debt, % of GDP B. Private borrowing, % of GDP 80 70 8 public private 6 60 4 50 40 2 30 0 20 -2 10 0 -4 92 94 96 98 00 02 04 06 08 10 12 92 94 96 98 00 02 04 06 08 10 12 Figure 2: Argentina’s external liabilities years 1992-3 it is less than 2% of GDP and it is close to zero on average. This is a consequence of heavy capital controls used to aid the fixed exchange 5 rate regime. These two facts motivate our assumption that households cannot borrow or lend. That is all the borrowing is done by the government. Modelling government directly opens a possibility for quantitative analysis of political risk as in Aguiar and Amador [2011]. A. Government budget, % of GDP B. External balance, % of GDP 8 6 20 fiscal deficit interest payments current account 15 4 10 2 5 0 0 -2 -5 -4 -10 92 94 96 98 00 02 04 06 08 10 12 92 94 96 98 00 02 04 06 08 10 12 Figure 3: Argentina’s budget and current account balances Further, reversals in budget balance can alone explain a large fraction of current account reversals in emerging economies.3 Figure 4 panel A plots the difference between the Argentina government’s revenues and expenditures. This does not include interest payments; this item is plotted separately. Thus at the time of 2001 default government deficit was more than 3% of GDP; at the same time the government paid 4% of GDP in the form of interest payments. Figure 4 panel B plots the trade and the current account balances. They increased 13.7% and 10.0% respectively between 2001 and 2002. Current account and trade balance in Argentina are strongly countercyclical. That is lending evaporates at distress times. 3 Ferrero [2010] shows that the U.S. current account can be well explained by the U.S. fiscal policy stance and demographical changes. 6 Government budget deficit, % of GDP 9 8 IMF calculations Implied by debt accumulation 7 6 5 4 3 2 1 0 1992 1994 1996 1998 2000 2002 Figure 4: Argentina’s budget and current account balances 3 3.1 Model Uncertainty, goods and technology The state zt ∈ S is a first-order Markov process. The state zt represents productivity of the country borrower. Let dF (zt+1 |zt ) denote the transition density from state zt into state zt+1 . We denote history of the state up to date t by z t = (z0 , z1 , ..., zt ). At each date a consumption good is traded. It is produced using a linear technology with labor being its single input: y(zt , lt ) = zt lt . 3.2 (1) Households A country-borrower is populated by a representative household with recursive preferences. Let (c(z t ), l(z t ), g(z t )) denote consumption, labor supply and public good consumption of the representative household in period t after history z t . The household evaluates utility from a plan {(c(z t ), l(z t ), g(z t )), ∀t, z t } 7 using a recursive utility: U (z t ) = W ((c(z t ), l(z t ), g(z t )), µ(U (z t+1 ))), (2) where W is an utility aggregator and µ ia a certainty equivalence function. For more details see Backus et al. [2004]. In this paper we study a special case with robust preferences: Z βb t t t t −θb U (z t+1 ) U (z ) = u(c(z ), l(z ), g(z )) − b ln e dF (zt+1 |zt ) . (3) θ The period utility function u(c, l, g) is strictly increasing and strictly concave in (c, −l, g). Parameter θb measures the degree of uncertainty aversion. But it can also be interpreted as a risk-aversion to intra-temporal wealth gambles as in Tallarini [2000]. We use the latter interpretation. This permits us to calibrate θb using error-detection probabilities as suggested by Hansen and Sargent [2007]. Households cannot borrow or lend and simply consume their own income in each period. Households take as given government’s current policy (τ, g, b0 ) that consists of labor income tax rate τ , public good provision g and borrowing b0 . Household’s problem then is: max u(c, l, g) : c = (1 − τ )zl. c,l (4) The optimality condition that we are going to exploit is: −ul (c, l, g)/uc (c, l, g) = (1 − τ )z. (5) Together with the budget constraint it allows us solving for the optimal labor supply policy l∗ (τ, z) given the tax rate t, the level of government spending g, and the productivity level z. We denote the indirect utility function of the household by w(τ, g, z): w(τ, g, z) ≡ u(c∗ (τ, g, z), l∗ (τ, g, z), g), (6) where (c∗ , l∗ ) denotes the optimal choice of the household. Finally, we can define the tax revenue function: T (τ, g, z) ≡ τ zl∗ (τ, g, z). (7) It depends on the government spending to the degree that the optimal labor supply does. 8 3.3 Government Penalty for defaulting is a permanent or a temporary exclusion from the international capital markets. When the government is in financial autarky it cannot borrow or lend. It chooses tax and spending policies (τ, g) to maximize the household’s welfare: Z βb t t t −θb A(z t+1 ) A(z ) = max [w(τ (z ), g(z ), zt )] − b ln e dF (zt+1 |zt ) (8a) τ (z t ),g(z t ) θ subject to a budget constraint: g(z t ) 6 T (τ (z t ), g(z t ), zt ). (8b) We also define a value function Ad for the country that is suffers economic penalties in addition to being banned from the international credit markets. Penalty comes in the form of reduced productivity d(zt ) 6 zt during the default period. Value function Ad must satisfy the following Bellman equation: Z βb −θb A(z t+1 ) t t t e dF (zt+1 |zt ) Ad (z ) = max [w(τ (z ), g(z ), d(zt ))] − b ln τ (z t ),g(z t ) θ (9) With access to financial markets the government chooses tax, spending and borrowing policies (τ, g, b0 ) to maximize household’s welfare: n V (z t ) = t max w(τ (z t ), g(z t ), zt ) τ (z ),g(z t ),b0 (z t ) Z o βb b t+1 t+1 e−θ max[V (z ,A(z )] dF (zt+1 |zt ) . (10a) − b ln θ subject to a budget constraint: g(z t ) + b(z t−1 ) 6 T (τ (z t ), g(z t ), zt ) + qb (b0 (z t ), zt )b0 (z t ). 3.4 (10b) International creditors International creditors are risk-neutral, two-period agents that rank alternative consumption streams according to: Z βc c t+1 c t t U (z ) = c(z ) − c ln e−θ c(z dF (zt+1 |zt ) . (11) θ 9 and discount future profits at gross rate R = 1/β c . Each creditor can supply one unit of funds to the country-borrower.4 Finally, we assume that creditors are more patient than the household in the country-borrower: β c > β b. (A) 3.5 Recursive representation and equilibrium We restrict our attention to Markov perfect equilibria. We postulate that government policies and repayment decision are functions of the current debt b(z t ) and current productivity level zt only. Let A(z) be the autarkic welfare when current productivity is z. It must satisfy the following Bellman equation: Z n o βb −θb A(z 0 ) 0 A(z) = max w(τ, g, z) − e dF (z |z) . (12) g6τ zl∗ (τ,g,z) θb Let V (b, z) be the optimal life-time utility of the household when the government chooses to repay its debt b and productivity is z. It must satisfy the following Bellman equation: Z βb −θb max[V (b0 ,z 0 ),A(z 0 )] 0 V (b, z) = max w(τ, g, z) − b e dF (z |z) (τ,g,b0 )∈B(b,z) θ (13a) where the budget set B(b, z) is defined as follows: n o B(b, z) = (τ, g, b0 ) : g + b 6 T (τ, g, z) + qb (b, z)b0 . (13b) Let 1r (b, z, z 0 ) ≡ 1(Ad (z 0 ) 6 V (b0∗ (b, z), z 0 )) denote a debt repayment indicator. Then creditor’s individual rationality implies: qb (b, z) = βc δ̃(b, z), where Z δ̃(b, z) = 1r (b, z, z 0 )dF̃ (z 0 |z) (14) (15) z 4 This assumption implies that the bond price is not a function of the amount a single creditor lends but only the gross debt outstanding. 10 is the next period’s ‘twisted’ probability of repayment. It is computed with respect to the ‘twisted’ probability measure F̃ : c 0 e−θ 1r (b,z,z ) dF (z 0 |z) . dF̃ (z |z) ≡ R −θc 1r (b,z,z0 ) e dF (z 0 |z) 0 (16) Given the above we can write: δ̃(b, z) = e−θc δ(b, z) , e−θc δ(b, z) + 1 − δ(b, z) (17) where δ(b, z) is the true, i.e. computed using F , probability of repayment. Definition. A recursive competitive equilibrium is debt price function qb , government’s tax, spending and borrowing policies (τ, g, b0 ) and household’s consumption and labor supply policies c, l such that: 1. (c, l) solve the household’s optimization problem (4); 2. (τ, g, b0 ) solve the government’s optimization problem (13a-13b); 3. qb satisfies the creditor’s individual rationality condition (14). 3.6 Simplified formulation The first-order optimality condition of the government is: ∂ ∂ wτ (τ, g, z) T (τ, g, z) − 1 = wg (τ, g, z) T (τ, g, z). ∂g ∂τ (18) The above equation implicitly defines the optimal public spending g ∗ (t, z).5 Using the optimal spending policy the optimization problem of the government can be reformulated in terms of (t, b0 ) alone. So, we can rewrite (13a) as follows: Z ∗ b 0 0 0 V (b, z) = max w(τ, g (τ, z), z) + β V (b , z )dF (z |z) (19) 0 (t,b ) z0 subject to the “indirect” budget constraint: g ∗ (τ, z) + b = T (t, g ∗ (t, z), z) + qb (b, z)b0 . 5 When the utility function is separable in the government’s spending as it will be assumed then (18) uniquely defines g ∗ . 11 For any choice of states (b, z) this is effectively a one-dimensional optimization problem that can be solved using standard numerical techniques. Let τ ∗ (b, z) and b0∗ (b, z) denote respectively the optimal tax and debt policies of the government. We conclude this section with a useful proposition. It shows that the government tax revenue, spending, and budget surplus are strictly monotone functions of t for all feasible tax rates. That is there is a unique way of implementing any desired level of government surplus. The implication of this result is that for any choice of (b, z) there is a unique solution to the government’s optimization problem.6 Proposition 1. Let z > 0. The following statements are true on the tax rate interval [0, (1 + η)−1 ]: A. The government tax revenue T (t, z) is a str. increasing function of t; B. The government spending g(t, z) is a str. decreasing function of t; C. The government budget balance S(t, z) is a str. increasing function of t. Proof. This can be shown directly by differentiation using formulas in the appendix A. Finally, because S(t, z) is a strictly monotone function of t and S(0, z) < 0 < S((1 + η)−1 , z), ∀z there exists unique taut (z) ∈ (0, (1 + η)−1 ), ∀z such that S(taut (z), z) = 0. The latter is the optimal level of taxation in autarky. Figure 5 plots the government budget surplus and the government spending for several choices of η: 0.5, 1.0, and 2.0. Panel A demonstrates the result stated in proposition 1. Panel B shows that the government spending is a decreasing function of the tax rate t. While the government tax revenue increases with the tax rate most of this revenue is saved by the government. It is optimal to do so to keep the desired balance between the declining private consumption and the public good consumption. 6 This also aids computations as the relation between government deficit and tax rate can be pre-computed. 12 A. Budget balance (surplus), S(τ,z) B. Government spending, g(t,z) 0.40 0.50 0.30 0.45 0.20 0.40 0.35 0.10 0.30 0.00 0.25 -0.10 0.20 -0.20 0.15 -0.30 0.10 η=0.5 η=1.0 η=2.0 -0.40 0.05 -0.50 0.00 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 tax rate, τ 0.1 0.2 0.3 0.4 0.5 0.6 0.7 tax rate, τ Figure 5: Government budget surplus and spending as functions of the tax rate. Parameters: γ = 2, χ = 0.5. 4 Results We illustrate our findings with a numerical example. Assume that the household’s one period utility is: l1+1/η + χln(g). (20) u(c, l, g) = ln c − a 1 + 1/η We assume that η = 1.00, a typical labor supply elasticity. We calibrate φ so that government spending is 15% of the country’s GDP. Productivity process is assumed to be an iid log-normally distributed random variable. The world interest rate is 5%. We would like the reader to regard this parameterization as an example rather than a rigorous calibration.7 We calibrate model mis-specification parameters to match the desired error-detection probability as suggested by Hansen and Sargent [2007]. The 7 Notice that most of the moments – output volatility, level of government spending, etc. – are endogenous and the model must be calibrated using the simulated method of moments. 13 parameter value moment βb 0.915-0.943 default probability is 0.75% βc 0.987 world interest rate is 6% γ 2.000 ‘typical’ value η 1.000 ‘typical’ value χ 0.300 gov. spending is 13.4% of GDP λ 0.282 Recovery probability d 0.850 Average output loss after default is 1-2% c θ 1.200 creditor’s edp is ≈0.40 b θ 0.800 borrower’s edp is ≈0.40 Table 1: Calibrated parameters error-detection probability (EDP) is the probability that the likelihood ratio model test errs. Let E1 denote occurrence of the type-I error – when a true model is incorrectly rejected. Let E2 denote occurrence of type-II error – when a wrong model is incorrectly accepted. Then EDP = 0.5[prob(E1 ) + prob(E2 )] ∈ [0, 0.5]. (21) Low EDP signifies that a researcher would err infrequently. This implies that models are easy to distinguish. If Fe is a ‘twisted’ version of F then: EDP = 0.5[prob(F accepted|Fe) + prob(Fe accepted|F )]. The above can be computed by Monte-Carlo simulation. Model π is accepted when it generates higher likelihood of a simulated sample than its alternative π̃. The larger the sample size the higher is EDP. We set sample size to 74, the length of the Argentina’s data. We choose θc and θb such that EDP is 0.40.8 That is if a random sample of length 74 is generated an econometrician would correctly discriminate the two models with probability of only 20%. The model solution is computed using collocation method for the Bellman equation. Collocation grid is a tensor product of 300 values of z and 500 values of b. The algorithm is stopped when a change in the implied interest rate, 1/q(b), is less than one tenth of a basis point. 8 A typical value used is 0.25. 14 A. Consumption, c B. Next period assets, b’ 0.80 0.20 45o 0.70 0.10 0.60 0.00 0.50 0.40 -0.05 zmin zmax 0.00 0.05 0.10 -0.10 -0.05 0.15 C. Government spending, g 0.00 0.05 0.10 0.15 D. Bond price, q 0.25 1.00 0.80 0.20 0.60 0.15 0.40 0.10 0.05 -0.05 0.20 0.00 0.05 0.10 0.00 -0.05 0.15 E. Tax rate, τ 0.00 0.05 0.10 0.15 0.10 0.15 F. Value fn, V 0.30 -70.0 -71.0 -72.0 0.25 -73.0 0.20 -74.0 -75.0 0.15 -76.0 -77.0 0.10 -0.05 0.00 0.05 0.10 -78.0 -0.05 0.15 0.00 0.05 Figure 6: Model solution when both creditors and the borrower are concerned about model mis-specification: θc = ln(1.2), θb = 0.8. Government’s assets, a, are measured on the x -axis. Each line corresponds to z = µz + iσz , i = −3, −2, ..., 3. Figure 6 demonstrates the model solution. The lowest asset level observed in equilibrium is -0.018. Panel A plots the household’s consumption level. It 15 is increasing in the productivity level z. But it is not a monotone function of the government’s assets b. But the region where non-monotonicity is observed is never reached in equilibrium. So, consumption is a monotone function of the government’s assets and the productivity level. The same is true about the public spending depicted in panel C. Both public and private consumption are driven by the tax rate that decreases monotonically as the government’s assets increase. When the government is in debt the tax rate varies little with the productivity level and it stays close to its maximum level of 19%. That is the government chooses to tax heavily to limit the decline in government spending. When the government has a sufficient amount of assets it taxes labor less, especially when the productivity level is low. Panel D plots the bond price function q(b). It is a monotone function of assets and largely reflects the shape of the cdf of the productivity shock z. While the bond price can potentially reach zero this does not happen in equilibrium as the government, “warned” with a ballooning borrowing cost, increases taxes sharply. Panel F plots the optimal value for the government when it decides to repay (increasing black lines) and when it decides to default (horizontal gray). The circles and the line that connects them marks the “default boundary.” These are points at which the government is indifferent between repaying and defaulting. Public spending declines slowly (note the scale) as government’s assets decline. That is the government chooses to tax labor more rather than to lower spending. In low productivity states such policy depresses output further and can lead to a crisis. Finally panel E plots an equilibrium bond price function. It largely reflects the distribution of the exogenous productivity state z. We now turn to the “twisted” distributions. The government-borrower in this setting behaves as if having ‘twisted’ beliefs. Its beliefs are determined endogenously in equilibrium as the ‘degree’ of twisting depends on the equilibrium outcomes: e−θb max[V (b,z),Ad (z)] . e−θb max[V (b,z),Ad (z)] dF (z) z dFeb (z|b) = R (22) Figure 7 plots ratios of distributions used by creditors and the borrower. Both 16 relative likelihood, dFtilde(z)/dF(z) 1.20 borrower creditor 1.15 1.10 1.05 1.00 0.95 0.90 0.85 0.80 0.90 0.95 1.00 1.05 1.10 productivity, z Figure 7: Ratios F̃ (z)/F (z) for θc = 1.20, θb = 0.80 twist their distributions pessimistically. For creditors adverse outcomes are those that lead to default. That is why creditors overestimate probability of outcomes in the left tail: F̃ c (z)/F (z) > 1 for z 6 0.961. The borrower twists the whole distribution. So, F̃ b (z)/F (z) > 1 for below the mean productivity levels, z 6 1, and F̃ b (z)/F (z) < 1 for above the mean productivity levels, z < 1. If we consider F̃ c (z)/F̃ b (z) there are three regions. This ratio is above one at the lower tail and the upper body of the distribution. This ratio is below one at the lower part of the distribution body.9 Disagreement leads to volatile returns because as the borrower’s wealth changes market power shifts between the borrower and creditors reflecting respective beliefs. With multiple assets endogenous disagreement would lead to frequent changes of market volume in select markets as in Routledge and Zin [2009]. But in our model we only have one asset; so, volatile demand driven by the disagreement must be reflected in more volatile prices and returns. Figure 8 demonstrates the economy’s dynamics along the path of low productivity realizations. The initial debt is assumed to be zero. As can be seen from the figure the debt is steadily increasing. The sovereign lending premium, that equals the “twisted” default probability here, stays zero until 9 See also figure 10 in the appendix for “twisted” distributions. 17 A. Debt B. Premium 0.050 0.030 0.025 0.000 0.020 -0.050 0.015 -0.100 0.010 -0.150 0.005 -0.200 0.000 0 2 4 6 8 10 0 C. Tax rate 2 4 6 8 10 8 10 D. Public spending 0.050 0.046 0.045 0.045 0.040 0.035 0.044 0.030 0.043 0.025 0.020 0.042 0.015 0.041 0.010 0.040 0.005 0.000 0.039 0 2 4 6 8 10 0 2 4 6 Figure 8: A run-up to a crisis: a sequence of shocks is {e−σz , e−σz , ...} period 5 when it jumps first above 0.01 then to 0.03. The tax rate is following the same path. An increased tax rate is needed to generate a larger surplus and so to slow-down debt accumulating. Not to discourage production the government balances higher taxes with lower public good provision. Betting on better productivity levels to realize in the near future the government is slowly led into default “zone.” While default does not occur along the assumed path of productivity shocks there is a significant probability of nonpayment starting with period t=5. Figure 9 presents simulated series. Default episodes correspond to periods where premium is negative. Panel A shows that government debt is very volatile. The government spends about a third of the time with no debt. 18 But government savings are decumulated quickly. So, a crisis-like situation can arise quickly. Bond returns are also volatile with perceived probability of default often rising above 4%. Despite this most of the time bond premium is zero. In the data the premium is always above zero even at times when the government’s debt is relatively low. One possibility to fix this could be to introduce disaster shocks like in Adam and Grill [2011] or a possibility of partial repayment. A. Government debt 0.05 0.04 0.03 0.02 0.01 0.00 -0.01 -0.02 50 100 150 200 250 300 350 400 450 500 350 400 450 500 B. Premium 0.05 0.04 0.03 0.02 0.01 0.00 -0.01 -0.02 50 100 150 200 250 300 Figure 9: Model simulation. Periods were the premium is negative are default periods. Table presents model moments. It contains four columns. The first column is the baseline parameterization in which neither the borrower nor creditors have concerns for model mis-specification. The second and the third columns turns on model misspecification for creditors and the borrower (one at a time) respectively. The fourth column features two-sided model mis19 specification. We vary the discount factor of the borrower to keep the default probably close to constant. We start with bond premia that is the focus of this paper. The mean bond premium does not vary much across specifications and is the range of 3.39 − 4.31%. That much lower than in the data but significant given how small is βc − βb relative to other studies. On the other hand, volatility of bond premium varies substantially across different specifications. When we turn on model mis-specification for creditors bond premium volatility goes up only marginally: from 6.51% to 7.12%. But when the borrower also has concerns about model misspecification then it increases to 8.21%. Interestingly that the two forms of model misspecification reinforce each other. This occurs for the following reason. If creditors have concerns for model misspecification country premium increases but the country borrower then contracts its borrowing. So, while the interest rate schedule becomes steeper high interest rates are not observed in equilibrium. Now suppose that the country-borrower also has concerns for model misspecification. Its expectations shift in the direction towards more pessimism. The value of staying in the market declines but the value of defaulting declines even more. So, the government willingly undertakes high interest loans because of fears of autarky. Thus, it is a robust borrower’s fear that leads him to borrow more, at higher rates, and not default longer than optimally. We will now discuss fiscal policy. As we turn on additional levels of model misspecification taxes become smaller and more volatile. That is as we introduce more pessimistic into the model we increase precautionary motives for the government. It taxes less on average and it also taxes more heavily in good times. That is fiscal policy becomes more “counter-cyclical.” The benefit of doing so is that the government can be more accommodating when productivity is low. This can be seen from the correlation ρz<µz (τ, z) that conditions on productivity being below its average. The government controls the level of pessimism present in equilibrium by not taxing the economy heavily in adverse states of the world.10 In other words, the government is managing expectations as in Karantounias [2013]. 10 Taxes still increase but not as much as they otherwise would. 20 θb = 0.00 θb = 0.00 θb = 0.80 θb = 0.80 θc = 0.00 θc = 1.20 θc = 0.00 θc = 1.20 βb = .943 βb = .938 βb = .938 βb = .915 µ[Rb ] 3.66 4.11 3.39 4.31 σ[Rb ] 6.51 7.12 7.95 8.21 ρ[Rb , y] -.71 -.69 -.75 -.75 µ(y) 0.82 0.82 0.83 0.82 σ(y) 7.29 7.30 7.29 7.29 µ(τ ) 0.22 0.33 0.12 0.11 σ(τ ) 0.40 0.48 0.62 0.65 ρ(τ, z) 0.41 0.28 0.81 0.89 ρz<µz (τ, z) 0.34 0.23 0.48 0.49 1 − µ(pr ) 0.75 0.75 0.75 0.75 b EDP 0.50 0.50 0.38 0.38 c EDP 0.50 0.41 0.50 0.41 Table 2: Model moments 5 Conclusions This paper builds a model of sovereign debt and default with an endogenously determined fiscal policy and two-sided concerns of model misspecification. Allowing for an endogenous fiscal policy enabled us to talk about fiscal deficit and government debt in pre-crisis periods. We showed that, like in the data, government deficit shrinks and even turns into a surplus, driving the current account out of red. Allowing for two-sided concern for model misspecification enabled us to improve the model’s predictions about sovereign bond spreads. While we enriched the model substantially computational complexity remains almost as that of simple exchange economies. This opens a possibility for modelling such first-order-of-importance factors as capital accumulation, political risk, and monetary policy. 21 References Klaus Adam and Michael Grill. Optimal sovereign debt default. Mannheim University manuscript, 2011. Mark Aguiar and Manuel Amador. Growth in the shadow of expropriation. Quarterly Journal of Economics, 126(2):651697, 2011. Mark Aguiar and Gita Gopinath. Defaultable debt, interest rates and the current account. Journal of International Economics, 69(1):64–83, 2006. S. Rao Aiyagari, Albert Marcet, Thomas J. Sargent, and Juha Seppala. Optimal taxation without state-contingent debt. Journal of Political Economy, 110:1220–1254, 2002. Cristina Arellano. Default risk and income fluctuations in emerging economies. The American Economic Review, 98(3):690–712, 2008. David K. Backus, Bryan R. Routledge, and Stanley E. Zin. Exotic Preferences for Macroeconomics, volume 19, pages 319–414. MIT Press, 2004. Andrea Ferrero. A structural decomposition of the U.S. trade balance: Productivity, demographics and fiscal policy. Journal of Monetary Economics, 57(4):478–490, 2010. Lars P. Hansen and Thomas J. Sargent. Robustness. Princeton University Press, 2007. Anastasios Karantounias. Managing pessimistic expectations and fiscal policy. Theoretical Economics, 8:193–231, 2013. Demian Pouzo. Optimal taxation with endogenous default under incomplete markets. UC Berkeley manuscript, 2013. Bryan R. Routledge and Stanley E. Zin. Model uncertainty and liquidity. Review of Economic Dynamics, 12:543–566, 2009. Thomas D. Tallarini. Risk-sensitive real business cycles. Journal of Monetary Economics, 45(3):507–532, 2000. and many more to be cited later... 22 A Derivations for CRRA-GHH preferences Assume the following utility specification: u(c, l, g) = 1 h l1+1/η i1−γ g 1−γ c−a . +χ 1−γ 1 + 1/η 1−γ (23) The solution to the household optimization problem is the labor supply function l∗ (t, z): l∗ (t, z) = ((1 − t)z/a)η . (24) The indirect (period) utility function of the household then is: a−(1−γ)η ((1 − t)z)(1−γ)(1+η) g 1−γ u((1 − t)l (t, z), l (t, z), g) = +χ . (1 − γ)(1 + η) 1−γ ∗ ∗ (25) The government’s FOCs allow to express the government spending g as a function of the tax rate t: g ∗ (t, z) = a−η z 1+η 1/γ χ (1 − t)1+η−1/γ (1 − (1 + η)t)1/γ . 1+η (26) Because government spending must be positive the above equation implies 1 that the tax rate cannot exceed 1+η . This bound decreases as the elasticity of labor supply η increases. Because the government spending and the labor supply are functions of (t, z) so is the government tax revenue, T (t, z), and primal surplus, S(t, z): T (t, z) ≡ tzl(t, z) = a−η z 1+η (1 − t)η−1 [1 − (1 + η)t)] S(t, z) ≡ tzl(t, z) − g(t, z) = a−η 1+η z 1+η (27) (28) h i (1 − t)η (1 + η)t − [χ(1 − t)γ−1 (1 − (1 + η)t)]1/γ . These results are used in the proof of proposition 1. 23 B Derivations for the special case with γ = 1 With preferences specified in (20) we get the following relations: l∗ (τ, g, z) = [(1 − τ )z/a]η , −η w(τ, g, z) = ln(a [(1 − τ )z] −η 1+η T (τ, g, z) = a z (29a) 1+η /(1 + η)) + φln(g), η τ (1 − τ ) . (29b) (29c) We now derive the optimal spending - tax relation: g ∗ (τ, z) = φa−η z 1+η (1 − τ − ητ )(1 − τ )η . (30) It is easy to show that g(τ, z) is positive and strictly decreasing function for all τ 6 1/(1 + η). That is there is a monotone relation between the desired spending and taxation levels. The total tax revenue is also a strictly increasing function of tax rate when τ 6 1/(1 + η). Fiscal deficit is: d(τ, z) ≡ g ∗ (τ, z) − T (τ, g ∗ (τ, z), z) = a−η z 1+η (1 − τ )η [φ − τ (1 + φ + φη)]. (31) It is a decreasing function of tax rate when τ 6 1/(1 + η). That is for any level of fiscal deficit there is a unique tax level that “implements” it. With access to financial markets the government chooses (τ, b0 ) subject to d(τ, z) = qb (b0 , z)b0 (b, z) − b. In financial autarky the optimal tax rate is determined from d(τ, z) = 0 as b0 = b = 0: τA∗ (z) = φ , 1 + φ + φη 24 ∀z. (32) 0.05 true borrower creditor 0.04 density dF(z) 0.04 0.04 0.03 0.03 0.02 0.01 0.01 0.01 0.00 0.90 0.95 1.00 1.05 1.10 productivity, z Figure 10: F̃ c (z), F̃ b (z) and F̃ c (z) for θc = 1.20, θb = 0.80 25