Government finance in the wake of currency crises

The paper addresses how governments actually pay for the fiscal costs of currency crises and what different financing methods imply for post-crisis inflation and depreciation, motivated by the observation that seigniorage does not rise sharply after crises as the classical view predicts (Section 1, Introduction). The classical view, associated with models in the Krugman (1979) tradition, holds that currency crises arise because governments print money to finance ongoing or prospective deficits, implying that seigniorage should rise significantly in the aftermath of a crisis. Table 1, reporting data for six recent currency crises (Thailand, Philippines, Mexico, Brazil, Turkey, and Korea), shows average seigniorage revenue before and after each crisis is of similar, modest magnitude – inconsistent with this prediction. This motivates the paper’s alternative approach: modeling a crisis as triggered by prospective government deficits and asking what other channels, besides seigniorage, governments actually use to pay for them.

The model gives the government access to three depreciation-related revenue sources – seigniorage, debt deflation, and implicit fiscal reform – rather than the seigniorage-only channel typical of the literature (Section 2). Standard models assume, for convenience, that seigniorage is the only depreciation-related revenue source. In reality, the authors argue, governments can also (i) reduce the dollar value of outstanding nonindexed debt through debt deflation, and (ii) benefit from an “implicit fiscal reform,” which arises when government expenditures denominated in local currency (transfers, and purchases of goods and services) are not raised at the rate of depreciation, so their dollar value declines. The paper’s government intertemporal budget constraint, equation (2.10), requires that the present value of any increase in transfers be financed by changes in the dollar value of: (i) tax revenues, (ii) government expenditures, (iii) CPI-indexed transfers, (iv) nonindexed debt, and (v) seigniorage.

These features are needed to generate a first-year rate of depreciation that is substantially larger than the first-year rate of inflation, matching the empirical wedge documented in the data (Section 4.4, “Deconstructing the Benchmark Model”). The authors show, through a sequence of numerical experiments, that a “simple textbook model” with no local-currency debt, all-tradable goods, flexible prices, and no distribution costs implies inflation and depreciation coincide exactly, at counterfactually high rates (23.3 percent inflation in the first year, Table 5(e)). Adding nominal debt lowers inflation modestly via debt deflation but still leaves inflation equal to depreciation. Only once nontradable goods with temporarily sticky prices are introduced does depreciation (34.2 percent in the first year) separate from inflation (20.3 percent); adding CPI-indexed transfers and, finally, distribution costs (a 50 percent pre-crisis distribution margin, calibrated to Burstein, Neves, and Rebelo (2002)) brings the model to its full benchmark specification, where first-year inflation is 12.7 percent against 46.5 percent depreciation.

Agents learn at time zero that the government will need to raise future transfers, and the fixed exchange rate is abandoned at the first point at which the domestic money supply has fallen by a fixed percentage of its initial level – a threshold that can occur well before any change in monetary policy actually takes effect (Section 3.1). Because the increase in transfers will not be financed by an explicit fiscal reform, the fixed exchange rate cannot be sustained indefinitely: if it were, the money supply would never change, the government could collect no seigniorage, and its intertemporal budget constraint could not hold. The authors solve for the equilibrium by fixing monetary-policy parameters (a threshold percentage, the timing of a later remonetization, and the post-remonetization one-time jump and growth rate of money), computing the resulting equilibrium paths, checking whether the government’s budget constraint holds, and iterating until it does. In the Korean calibration, this produces a speculative attack four months after the initial news, consistent with Korea’s crisis following Thailand’s by about four months, and, as in Sargent and Wallace (1981), inflation rises in anticipation of the later monetary expansion, before that expansion is actually implemented.

In the Korean benchmark calibration, first-year inflation is 12.7 percent and long-run inflation only 2.0 percent, while first-year depreciation is 46.5 percent – more than three times higher – with seigniorage financing only about 2.0 percent of pre-crisis GDP, or less than 10 percent of total depreciation-related revenue (Section 4.2, Table 5(a)). The dominant source of depreciation-related revenue in the model is the fall in the dollar value of transfers, at 11.1 percent of pre-crisis GDP, reflecting the wedge that opens up between the CPI and the exchange rate immediately after the crisis. The authors conclude that the government can meet most of its financing needs through this implicit fiscal reform channel, consistent with the empirical observation that many large devaluations combine low inflation with a modest role for seigniorage.

Removing either nominal (nonindexed) debt or CPI-indexed transfers from the benchmark, while requiring the government to make up the resulting revenue shortfall through higher money growth, substantially raises both inflation and depreciation (Section 4.3, “Implications of Alternative Financing Scenarios,” Table 5(b)-(d)). Eliminating nonindexed debt raises first-year inflation from 12.7 to 16.4 percent and long-run inflation from 2 to 6 percent. Making all transfers dollar-indexed (removing the implicit-reform channel on transfers) is more consequential: first-year inflation rises to 22.5 percent, long-run inflation to 12.7 percent, and first-year depreciation climbs from 46.5 to 60.3 percent. Removing both channels simultaneously produces the most extreme outcome: 29.2 percent first-year inflation, 20.1 percent long-run inflation, and depreciation near 70 percent. The authors conclude that post-crisis inflation and depreciation depend critically on which depreciation-related revenue sources are available to a given government, which is consistent with the substantial heterogeneity in post-crisis inflation and devaluation rates actually observed across countries.

None of the three countries had, by the end of the available data, fully financed the fiscal costs of its crisis; in no case was seigniorage the dominant financing source, and in all three, debt deflation exceeded seigniorage (Section 5, summary of findings). The authors estimate that Korea, Mexico, and Turkey had financed roughly 43, 52, and 54 percent of their respective crisis costs (Section 5, restated more precisely in the country summaries as 42.9, 52.2, and 53.8 percent). Seigniorage’s share of net crisis costs ranged from about 7 percent (Mexico) to about 10 percent (Turkey). Debt deflation’s share ranged from a low of about 7 percent of total cost in Mexico to a high of about 38 percent in Turkey. In all three countries there was a substantial post-crisis decline in the dollar value of tax revenues, but this was significantly offset by a decline in the dollar value of government purchases; and large declines in the dollar value of transfers were, in Korea and Mexico, the single most important financing source.

The authors estimate that the fiscal cost of Korea’s banking-sector bailout was about 24.0 percent of 1997 GDP (114.4 billion dollars), that the government had, as of the available data, raised 9.6 percent of GDP (42.9 percent of total costs) in depreciation-related revenue, and that transfers – not seigniorage – were the dominant financing channel (Section 5.1). Combining the bailout cost with post-crisis-recession revenue losses (10.2 percent of GDP) and netting out explicit fiscal reforms already enacted (11.9 percent of GDP) and expected (8.8 percent of GDP), the authors calculate the government needed to raise about 22.3 percent of GDP from depreciation-related revenue. Of the 9.6 percent of GDP raised to date, 7.5 percentage points came from the implicit decline in the dollar value of transfers, 3.5 points from debt deflation, and 1.7 points from seigniorage, partly offset by a −3.1 point decline in revenue net of government purchases. To construct these estimates the authors compare actual outcomes for seigniorage, tax revenue, government purchases, and transfers against counterfactual “no-crisis” trend paths estimated from pre-crisis data (1980-97), decomposing each deviation into output, relative-price, and residual “explicit reform” components.

Mexico had financed 52.2 percent of its crisis costs (12.5 percent of GDP) by the sample’s end, primarily through a 23.4-percentage-point decline in the dollar value of transfers, and the authors project that, absent evidence of further fiscal reforms, most of the remaining costs are likely to be financed by seigniorage (Section 5.2). The fiscal cost of Mexico’s banking-sector bailout was estimated at 15.0 percent of 1994 GDP; combined with recession-induced revenue losses (6.5 percent of GDP) and net explicit fiscal reforms of −2.5 percent of GDP (i.e., a net loosening), the government needed to raise 23.9 percent of GDP in depreciation-related revenue. Debt deflation and seigniorage each contributed only 1.7 percent of GDP to date, reflecting Mexico’s relatively modest stock of peso-denominated debt, while the decline in transfers was offset by a −14.3 percentage-point decline in revenue net of purchases. The authors project that if money growth and depreciation continued at their (pre-crisis) counterfactual rates, seigniorage could raise a further 12.5 percent of pre-crisis GDP – slightly more than needed to cover the remaining 11.4 percent of GDP – implying, per the model, that Mexican inflation would end up considerably higher than it would have been with larger implicit reforms or more initial domestic debt.

Turkey financed 53.8 percent of its crisis costs (10.3 percent of GDP) roughly twice as fast as Korea or Mexico, relying much more heavily on debt deflation and seigniorage rather than on implicit reductions in transfers and purchases (Section 5.3). The fiscal cost of Turkey’s banking-sector bailout was estimated at 18.2 percent of 2000 GDP; combined with recession-induced revenue losses of 4.4 percent of GDP, and net of already-enacted explicit reforms of 3.3 percent of GDP, the government needed to raise 19.2 percent of GDP in depreciation-related revenue. Of the 10.3 percent of GDP raised, 7.3 percentage points came from debt deflation – reflecting Turkey’s much larger stock of domestic debt outstanding relative to GDP (28.5 percent, versus 16.3 percent for Korea) – and 1.8 points from seigniorage, while the net decline in transfers minus purchases contributed only about 1.2 points. The authors attribute the two-year (versus five- and eight-year) financing horizon and the different revenue mix to Turkey’s greater reliance on debt deflation and seigniorage rather than implicit fiscal reform through transfers and purchases.

The authors conclude that models assuming seigniorage is the sole depreciation-related revenue source give misleading predictions about post-crisis inflation and depreciation, whereas a model incorporating the fuller menu of revenue sources they identify can account for the empirically common combination of high depreciation and low inflation – but the paper does not explain why different governments chose different financing strategies (Section 6, Conclusion). The case studies show that different countries relied on different channels – debt deflation mattered more in Turkey than in Korea or Mexico, while the decline in the dollar value of transfers mattered far more in Mexico – and the authors explicitly identify understanding the political-economy determinants of these choices as “an important topic for future research,” rather than something their model addresses.