A. The Genesis of Hyperinflation

A series of events in 1990 set the stage for the gradual disintegration of the political system and the loss of control over economic and financial management. First, on April 24, 1990, President Mobutu announced an upcoming process of democratization and the establishment of a multi-party system within a 12-month period. Second, political opposition gathered forces and organized widespread demonstrations to demand an acceleration of the democratization process. Third, with a view to appeasing demonstrators, the authorities awarded unsustainably large increases in government wages. And, fourth, the government’s financial difficulties were compounded by a drop in mining sector revenue.

In late 1989, Zaïre was swept by the winds of democratization that arose in the wake of the collapse of the former Soviet bloc. As a means to quell domestic dissent and foreign criticism, President Mobutu announced a forthcoming revision of the constitution, and the organization of multiparty elections. A new government was named to conduct the transition and address worsening economic conditions, but political parties were not liberalized, pending an unspecified selection process that would limit the number of political parties to three, to avoid stirring up ethnic tensions. Even though the opposition leader, Mr. Tshisekedi, was released from house arrest, demonstrations by his supporters were violently suppressed. On May 11, 1990, security forces cracked down on protesting students in the southern city of Lubumbashi (capital of the Shaba province, formerly known as Katanga), with a death toll of perhaps 100. Zaïre’s refusal to allow an independent inquiry into the massacre of Lubumbashi ultimately led to the suspension of most external aid.⁵

Faced with growing discontent, the authorities granted government employees large wage increases that nearly trebled the wage bill by October 1990 relative to the 1989 average. At the same time, government revenue began falling, reflecting a drop in mining taxes and dividends. Total government revenue fell from the equivalent of US$900 million in 1989 to less than US$800 million in 1990, wholly as a result of lower tax and royalty payments from the main copper mining concern, Gécamines (Générate des Carrières et des Mines). In turn, Gécamines’ difficulties stemmed partly from declining world market prices, but above all from a 12 percent fall in copper output resulting from a cave-in at a major mining site (Kamoto).

Overall, the government deficit on a commitment basis increased sharply in 1990, to the equivalent of US$l.l billion (13.1 percent of GDP), compared with US$0.8 billion in 1989 (9.3 percent of GDP). Apart from the buildup of external payments arrears, and excluding the disbursement of project assistance, the deficit was entirely financed by the domestic banking system. Domestic bank financing covered half of nondebt government expenditure in 1990, and accounted for the bulk of the 187 percent increase in broad money. By late 1990, inflation and currency depreciation reached monthly rates in excess of 40 percent.

In December 1990, the government allowed the establishment of political parties and lifted controls over the press. Demonstrations by the opposition and fierce repression by security forces continued in the spring of 1991. Under the circumstances, the authorities were not in a position to organize national elections; instead, the President convened a national conference, to draft a new constitution and prepare the ground for legislative elections. When it opened, on August 7, 1991, the national conference had about 2,800 participants, representing 225 political groups. Soon, however, the opposition called for a boycott, and President Mobutu suspended the national conference on August 15, 1991. In the midst of a worsening economic and financial situation, a mutiny erupted in September 1991 in the army, with demands for higher wages and the payment of back pay. As the mutiny generalized into widespread looting throughout the country, Belgium and France organized the evacuation of some 10,000 expatriates, who in most cases had to abandon their businesses and properties. The outbreak of violence and large-scale looting resulted in extensive destruction of basic infrastructures, which accelerated the fall in economic activity.

The opening of the national conference marked the beginning of an institutional stalemate. Yielding to calls from the opposition, President Mobutu appointed Mr. Tshisekedi as Prime Minister in late September 1991, but dismissed him within a few weeks. Reflecting a dramatic fall in economic activity and recurrent strikes in the civil service, revenue collection was halved, to less than US$400 million in 1991, but current spending continued to increase. Broad money growth in 1991—at 2,230 percent—again reflected entirely the government’s recourse to domestic bank financing. Monthly inflation averaged 37 percent over the year as a whole, with a peak of 114 percent in November 1991. By then, it was clear that the economy had plunged into a spiral of hyperinflation, currency depreciation, and falling output.

B. Economic Tailspin, 1992-93

Throughout 1992 and 1993, Zaïre’s political and economic situation continued to worsen at an alarming pace. The President suspended the national conference on several occasions, which led to widespread unrest and rioting. When it was reconvened in April 1992, the conference declared itself “sovereign,” with enforceable decisions. On August 15, 1992, the national conference elected Mr. Tshisekedi as Prime Minister, and the President promulgated his nomination on August 19, but the government remained largely powerless. As inflation eroded the value of the currency, the Bank of Zaïre issued new banknotes with a face value of Z 5 million in December 1992 (worth about US$2), which were declared illegal by Prime Minister Tshisekedi. Also in December 1992, the national conference concluded its work by establishing itself as the High Council of the Republic (Haut Conseil de la République, or HCR) to act as a transition parliament pending the organization of national elections. In the next week, President Mobutu ordered the permanent secretaries to run their respective ministries in lieu of the cabinet named by Prime Minister Tshisekedi.

The Treasury endeavored to settle its commitments, including wages and salaries, with the newly issued banknotes, but most recipients refused to accept the new notes. In December 1992, and again on several occasions during January 1993, riots and looting broke out throughout the country, and soldiers demonstrated, demanding the payment of salary arrears. An estimated 1,000 people were killed in clashes on January 22, 1993.

During 1993, the relations between President Mobutu and Prime Minister Tshisekedi continued to deteriorate. On January 29, the President issued an ordinance removing Prime Minister Tshisekedi from office, but he refused to step down. The President charged the Prime Minister with high treason, while the national conference stated its intention to impeach President Mobutu. In February, the Bank of Zaïre announced the issuance of Z 10 million banknotes (about US$4); at the same time, a 15-ton shipment of Z 5 million notes was seized in Belgium at Mr. Tshisekedi’s request.

In March 1993, the President named another prime minister, and convened a “conclave” with a view to resolving the political stalemate and drafting a new constitution. However, members of the HCR declined to participate in the conclave, which comprised mostly former parliamentarians allied to President Mobutu. During most of the year, Zaïre had two prime ministers and two parliaments, and no working government. Over January-September 1993, inflation slowed to about 13 percent on average per month, mainly reflecting the Treasury’s inability to effect payments with the new currency notes.

On October 22, 1993, the Bank of Zaïre implemented a currency exchange, in which the zaïre was replaced by the new zaïre at a rate of NZ 1 = Z 3 million, with an initial parity of NZ 3 per U.S. dollar. The new notes were again boycotted by the opposition, and soldiers refused to accept them in salary payments, which gave rise to renewed rioting. The central bank tried to slow down the depreciation of the new currency in the interbank market, but by year-end the exchange rate of the new zaïre had depreciated to NZ 102 per U.S. dollar in the parallel market, and the differential with the interbank market reached 191 percent. Monthly inflation surged to about 250 percent in November and December 1993, and 190 percent in January 1994—yielding cumulative price increases that exceeded 3,300 percent during the three-month period ending in January 1994.

Beginning in 1992, and until early 1995, domestic bank deposits traded at a discount against currency notes, a phenomenon referred to as the domestic money exchange rate. In parallel with the acceleration of inflation and the extraordinary increase in velocity, shortages of currency notes at the central bank at times made the conversion of bank deposits into cash next to impossible. Given the high costs of printing currency, the central bank often earmarked its stocks of banknotes for the government’s use (and its own as well), and refused to convert commercial banks’ reserves into cash; in turn, depositors were unable to mobilize their bank balances, except through transfers or checks at a steep discount. The money exchange rate, which reports the value of currency notes relative to bank deposits, surged from about 2 in the spring of 1992 to 12 in early 1993, and peaked at 170 in June 1993 (see IMF [1996a], p. 87).

The velocity of circulation of broad money rose from about 17 in 1989-90 to nearly 30 in 1991-92, and dropped to 15 in 1993 before surging to 60 in 1994 (Table 1). However, the 1993 drop was only apparent insofar as bank deposits, which accounted for up to 73 percent of broad money, were nearly worthless (given the steep discount of bank deposits over currency). Currency substitution presumably continued to develop at a rapid pace during this period: by end-1993, the stock of new zaïres in circulation was worth only US$46 million at the parallel market exchange rate, down from US$158 million at end-1992 and more than US$300 million at end-1989. Rough estimates suggest that the circulation of foreign banknotes—primarily U.S. dollars, but also CFA francs in the provinces next to BEAC member countries and Zambian kwachas in southern Zaïre—probably rose to the equivalent of US$300-400 million in Zaïre. Nor was the co-circulation of currencies limited exclusively to foreign banknotes: in the aftermath of the currency exchange, and in response to the boycott, old zaïres continued to circulate in the Western Kasaï province at a rate that gradually appreciated against the U.S. dollar, from about Z 20 million per U.S. dollar at end- 1993 to Z 15-16 million more recently.

At the time of the currency reform, some Z 460 trillion in old zaïre notes (slightly more than one-third of total currency issue) was not exchanged for new notes, or the equivalent of US$30 million at the present exchange rate for the old zaïres (see Bank of Zaïre, Rapport Annuel 1994, p. 174).

As the country’s infrastructure crumbled and became unreliable, Zaïrian entrepreneurs developed in partnership with U.S. companies an efficient cellular telephone network. Among others, currency traders became regular users of the cellular network. Instant and reliable communication was an important tool to ensure the timely transmittal of information in the parallel exchange market. In particular, currency traders in Zaïre were on the watch for the sudden unloading of large quantities of banknotes, which unvariably exerted an immediate impact on the parallel exchange rate (“Corando effect”).⁶

Anecdotal evidence suggests that, from 1990 onward, prices in Zaïre were increasingly set in foreign currency (U.S. dollars or Belgian francs), although small-scale transactions generally continued to be carried out in domestic currency. As wages and salaries in the formal economy were settled in domestic currency, wage earners adopted substitution practices to avoid relying on the new Zaïre as a store of value. Hence, foreign exchange bureaus and street money changers were particularly active, handling transactions on a continuous basis—from a few dollars to several thousand dollars. While some barter mechanisms developed as well (for example, companies provided their employees with transportation and housing services, or individuals paid teachers with basic food items), overall, barter remained limited in scale.

Reflecting a combination of several factors, the rates of inflation and currency depreciation were highly unstable. As a rule, Zaïre’s currency depreciated in sudden and rapid steps, when new notes were made available to the Treasury on a large scale, or when impending deliveries were expected. However, there was often a pause in the depreciation, or even a slight appreciation of the currency, toward the end of each month, as companies built up stocks of banknotes to meet their wage payments and tax liabilities.

C. Steady Hyperinflation, 1994-96

On January 14, 1994, President Mobutu dismissed the two rival governments and launched consultations to resolve the political stalemate. Although the radical opposition refused to negotiate, a number of moderate members of the opposition accepted the merging of the national conference and the presidential conclave into a transition parliament (Haut Conseil de la République - Parlement de Transition, or HCR-PT) that comprised some 800 members. In March 1994, the Parliament approved a formal arrangement providing for a 15-month transition, through July 1995, during which the prime minister would be chosen in consultations with political parties from outside the President’s party. Mr. Kengo wa Dondo, a leader from the moderate opposition, was elected Zaïre’s Prime Minister by the Parliament on June 14, and his nomination was ratified by President Mobutu soon thereafter.

The task of Prime Minister Kengo as set out by the HCR-PT was to lead the country to national elections by end-June 1995, but in the event this proved unrealistic (on July 7, 1995, the transition period was extended by another two years, through July 1997). In the meantime, the government endeavored to regain control over economic and financial management, and to reestablish normal relations with Zaïre’s foreign partners. The strengthening of financial management had two main components: first, the restoration of effective control over the issuance of currency, and, second, the balancing of budgetary operations.

The extraordinary inflation episode of late 1993 had given rise to widespread rumors of disorderly currency issues, including possibly the printing of counterfeit notes. It is doubtful that the Zaïrian currency was counterfeited on a large scale at the time, considering that seigniorage gains were rather low (because the Bank of Zaïre often delayed the introduction of high denomination notes). However, “legal” banknotes were often issued during 1993-94 without proper accounting in the balance sheet of the central bank, which was facilitated by a shift to a less reputable foreign supplier of currency notes—ostensibly as a means of lowering printing costs.⁷

Soon after coming into office, Prime Minister Kengo suspended the governor of the Bank of Zaïre in July 1994, and named his deputy as acting governor. However, the governor refused to step down until the revocation decree was signed by the President, in November. A new governor was named at the Bank of Zaïre in January 1995, or six months after the prime minister’s decision to suspend the former governor. During the first half of 1995, the Bank of Zaïre reverted to its traditional supplier of banknotes and security paper, and endeavored to recover stocks of banknotes and printing plates from the less reputable company.

Money growth slowed down gradually in 1994, from about 230 percent in the first and second quarters to 155 percent in the third quarter and 100 percent in the fourth quarter of the year. Inflation remained surprisingly high, however, in excess of 50 percent a month on average during January-September 1994, before declining to about 25 percent in October and November, and less than 10 percent in December.

With the progressive reestablishment of control over central bank operations in early 1995, the authorities turned their attention to fiscal management. Measures were taken to improve government revenue collection, and the Treasury was instructed to tighten expenditure control procedures and limit actual spending within available resources. As a result, government revenue rose to some US$150 million during the first half of 1995, or 10 percent more than the total collections for 1994. Expenditure sequestration, which involved inter alia the accumulation of wage payment arrears, resulted in a buildup by the Treasury of surplus balances with the Bank of Zaïre, and inflation slowed to a monthly average of 10 percent during the first half of the year. Residual inflationary pressures stemmed from a deficit in central bank operations, mainly reflecting the large costs associated with currency printing and payments on short-term external debt obligations, as well as a strong expansion in central bank refinancing that accommodated the weakening position of several commercial banks.

By mid- 1995, the disinflation gains gradually ground to a halt. Spending by the Treasury rose somewhat, and the deficit of central bank operations remained large. Inflation was subdued during July-August, but surged in the latter part of the year. Beginning in September, the authorities yielded to political pressures and started spending the balances accumulated with the central bank, to settle the allowances of parliamentarians, clear government wage arrears, and launch urgent public works such as road repair. With the monetization of government deposits, inflation surged from a monthly average of 13½ percent during the third quarter of 1995 to 23 percent during the last quarter, and to 34 percent in January 1996 alone. In turn, government revenue collection weakened to US$55 million in the last quarter of 1995, as against some US$90 million in the second and third quarters.

Following its January 1996 surge, monthly inflation hovered in the range of 10-20 percent, while government revenue remained on the order of US$20 million a month. At least until late 1996, the precarious political modus vivendi found in 1994 appeared to be holding. Zaïre remained on the brink of hyperinflation, but positive real GDP growth was expected for the first time since 1988. Although it is doubtful that hyperinflation can be “stabilized” durably, as the process is inherently highly unstable, economic and financial collapse had been avoided since the extraordinary inflation episode of late 1993, owing to the imposition of stricter controls over public finances and central bank operations.

The war in Kivu, in which Tutsi rebels seized major cities in eastern Zaïre in the latter part of, brought strong pressures on the government of Prime Minister Kengo and renewed calls by the radical opposition for the nomination of Mr. Tshisekedi as prime minister. After undergoing cancer treatment in August 1996, President Mobutu convalesced in Europe for more than three months before returning to Kinshasa with the aim of retaking control over the country. The crisis in the Great Lakes region and the speed at which rebels routed the Zaïrian army highlighted the extraordinary weakness of government authority.

As of end-1996, in addition to losing control over a sizable portion of Zaïre’s territory, the government’s hold over the financial situation appeared to have weakened as well. Reflecting a surge in government spending, monthly inflation accelerated to 24 percent during October-December, while the money exchange rate, which had nearly disappeared since early 1995, rose to about 2. The inflation rate for the 12-month period ending December 1996 reached 657 percent, equivalent to a monthly average of 18½ percent.

As indicated above, throughout 1995 and 1996, the Bank of Zaïre remained the main impetus behind inflationary pressures: its deficit mostly reflected the cost of printing currency and the settlement of short-term external liabilities, including arrears vis-à-vis the foreign supplier of banknotes and security paper. The average cost of banknotes issued by the Bank of Zaïre was relatively low (on the order of US$100 per thousand, including transportation charges), but volumes were considerable. Thus, the Bank of Zaïre issued some 830 million new banknotes in 1995 (in part out of the fear that the introduction of higher denomination notes would elicit adverse political reactions, as in 1992 and 1993) at an aggregate cost of about US$80 million. During this period, the currency stock increased by some NZ 1½ trillion,⁸ or the equivalent of US$180 million (converted monthly at the parallel exchange rate). In effect, nearly half of the currency newly issued by the Bank of Zaïre in recent years was used to acquire foreign exchange in order to pay for the issuance of currency!

From 1989 to 1994, the Zaïrian economy contracted cumulatively by some 40 percent, and consumer prices rose by a factor of 21 million; government revenue collections fell from nearly US$900 million in the late 1980s to US$138 million in 1994. During 1995-96, the contraction in economic activity bottomed out, and even showed some signs of reversal, while prices rose by a factor of about 35. But even with sustained progress toward stabilization of the economy, the task ahead to rebuild the country is enormous, and the challenge of restoring the authority of the state is perhaps even greater. By end-1996, Zaïre’s infrastructure had fallen into disrepair: most roads, railways, and rivers had become impracticable; the government had all but stopped providing health and education services; and many public enterprises had ceased operating. The informal sector showed an amazing resilience, which partly buffeted the plummeting activity in the formal economy. But the development of “survival reflexes”—with greater autonomy of provinces and generalized self-subsistence, in particular—also translated into “noncivic” behavior, including the extension of corruption and the ransacking of the population by unpaid soldiers. Hyperinflation in Zaïre in the 1990s thus left a trail of deleterious consequences on the fabric of society, which will take many years to remedy.⁹

A. Money Supply and Demand

Money supply is a direct function of the government deficit:

This is an identity, since the government sector (inclusive of the central bank) is by convention the sole source of money creation. Thus, any increase in the stock of broad money necessarily reflects a deficit in government operations insofar as there are no other financing sources, either domestic or external.

The demand for real money balances in domestic currency is:

Equation (2) takes the usual form, as in Cagan’s model, where the demand for real balances in domestic currency depends on expected inflation (α > 0), but also on the level of real non-mining revenue (β > 0); A is a scaling constant. Nonmining revenue is included as a proxy for the demand of transaction balances in domestic currency (mining revenue is not taken into account, because it is paid directly in foreign currency).¹² As inflation reaches very high levels, the local currency loses its usefulness, both as a store of value and as a medium of exchange, and even as a unit of account. As a result, economic operators tend to reduce their cash holdings as much as possible and substitute foreign exchange; even common transactions are increasingly conducted in foreign currency notes. Yet, the demand for domestic currency does not fall to zero as long as the government continues to collect revenue: even if inflationary expectations were to become infinite, taxpayers would still need to acquire domestic currency in order to settle their obligations to the Treasury—and thus real money demand is unlikely to drop to zero unless government revenue collections collapse entirely.

Applying a log transformation to equation (2) yields:

Equation (3) is expressed in terms of prices, which is the adjustment variable that clears the money market. The superscript for M_t has been dropped, since equilibrium in the money market implies that money supply equals money demand. Ceteris paribus, any increase in money supply translates into higher prices, by a factor of 1/ (1-β). Equation (3), however, is not a reduced form because M_t is itself a function of RO_t

B. The Tanzi Effect

The Tanzi effect refers to the erosion of the tax base by inflation.¹³ There is unavoidably a lag between the time tax payments are assessed and the time they are collected by the Treasury. In the case of indirect, domestically based taxes (such as turnover taxes), the collection lag is generally four to six weeks; for direct taxes, the lag is much longer, and may reach well over a year for personal income taxes. Even in the case of customs duties, where the assessed base is denominated in foreign currency, taxpayers are generally allowed several weeks to make payments to the Treasury. As regards nontax revenue such as licenses and fees, the amounts due are normally set in local currency terms, which tend to make their value insignificant under conditions of hyperinflation.

In Zaïre, where the bulk of government revenue consists of indirect taxes, the average collection lag may be on the order of one to three months. The Tanzi effect on government revenue is thus quite large: with inflation running at 10 percent a month, real revenue collection drops by 9.1 percent if the collection lag is one month, or 17.4 percent if the lag is two months (relative to the case with no inflation). As inflation reaches 50 percent a month, the drop in real revenue collection amounts to 33.3 percent or 55.6 percent, respectively.¹⁴ Under such circumstances, taxpayers will always endeavor to delay the settlement of tax obligations; moreover, tax and customs agents will often use their prerogatives to allow delayed payments and share some of the taxpayers’ gains.

In the general case, various taxes are collected with lags ranging from zero to n months; non-mining government revenue is thus specified as:

While an arithmetic specification is more commonly used, the geometric formulation is better suited to a hyperinflation environment; it also provides computational convenience, as it can be linearized by a log transformation. Let k be the average collection lag; as short-term fluctuations in output are not considered, equation (4) may be simplified as follows:

This formulation has the advantage of separating price effects, while eliminating real income from the distributed lag structure (there are no reliable monthly indicators of activity in Zaïre, and the index used for estimation purposes is simply a geometric interpolation of real GDP).

C. Modeling the Hyperinflation Process

In the standard model, the stability of the inflation process depends entirely on the behavioral coefficients for expected inflation: the inflationary process is stable “when money demand responds little to expected inflation and when expectations adjust sluggishly.” Under rational expectations, “there is typically an infinity of solutions for the price path, with only one of them not implying unstable price behavior” (Dornbush and Fischer [1986], Section II).

The introduction of the Tanzi effect in the base model makes things worse, relative to the case where taxes are not considered, insofar as inflation lowers government revenue and thus enlarges the deficit in real terms. When real government spending is set at a level in excess of current revenue, inflation increases, ceteris paribus, which reinforces the spiraling impact of inflationary expectations. At the same time, however, the inclusion of real nonmining government revenue in the money demand function acts as a brake on this vicious spiral.

Many historical precedents testify that hyperinflation is inherently highly unstable, and quickly leads to total collapse and forced stabilization.¹⁵ The famous Weimar episode in Germany lasted only 15 months, and the record Hungarian hyperinflation of 1945-46, less than a year. Developments in Zaïre since 1990, however, show that hyperinflation can be sustained and total collapse avoided for many years. A number of Latin American countries have also experienced triple- or quadruple-digit inflation rates for several consecutive years (notably Brazil, 1987-94, and Nicaragua, 1985-90). A key aspect of these long-lasting cases of extraordinary inflation was the high variability of inflation, as a consequence of large fluctuations in real government spending (see Végh [1992]).

An important prerequisite for modeling the hyperinflation process, therefore, is to treat government spending as a variable rather than as a fixed parameter.¹⁶ Highly variable real government spending and unstable inflation are common features of all hyperinflation episodes, and should be seen as intrinsic elements of the hyperinflation process. When inflation reaches very high levels and the potential gains from seigniorage tend to disappear, or when no foreign exchange is available to print new notes and adapting the face values of banknotes no longer seems a viable option, the government is compelled to reduce spending temporarily. With government spending sharply down, inflation drops quickly, which creates the opportunity for a new outburst of government spending.¹⁷

There is no straightforward way to reduce equations (1), (3), and (5), because of the linear structure of the money supply equation. Combining these equations yields:

Two alternative specifications are used for inflationary expectations. Under rational expectations (RE), there is no uncertainty and π^e is equal to the actual rate of inflation (implying a perfect “Corando effect”), while under adaptive expectations (AE), Δπ^e is a function of the discrepancy between actual and expected inflation in the previous month:

The adaptive expectation specification is equivalent to a distributed lag of past inflation rates, which offers some parallelism with the specification of the Tanzi effect:

Given the inclusion of government revenue in the money demand function and the Tanzi effect, it may be difficult in practice to distinguish the RE and AE specifications, since in both cases the price equation (3) will involve distributed lags of Log P_t (or ΔLog P_t).

Substituting (8) into (3) is most straightforward in the RE case:

or:

In the AE case, π^e is eliminated by subtracting (1-λ) Log P_t-1 from both sides in equation (3):

or:

Neither equation 10 nor equation 12 can be reduced to a difference equation in terms of prices, insofar as distributed lags of P_t enter into RO_t, and thus into M_t as well. As a rule, an important condition for the stability of the inflation process is that the first coefficient in equation 3 (α/(1-β)) should be less than unity (i.e., α +β < 1). If this condition is not met, inflationary expectations could exert an explosive self-fulfilling effect, in which higher prices generate ever higher inflationary expectations. Yet, this condition is neither necessary nor sufficient to ensure stability. This is because the second element in equation 3 tends to reinforce inflationary pressures (especially if β is low—that is, if money demand is little sensitive to changes to government revenue collection, and depends almost entirely on inflationary expectations), while the third element has the reverse effect. In other words, the inclusion of government revenue in the money demand function, together with the specification of the Tanzi effect, partly neutralizes the impact of expectations on inflation—expectations essentially affect the speed of adjustment, rather than its path.

A. Government Nonmining Revenue

Direct estimates of the Tanzi effect on nominal government nonmining revenue based on first differences do not yield significant results. Alternatively, equation (5) above may be transformed to express government nonmining revenue in real terms:

Insofar as all τ_i are positive while their sum should be close to 1, the distributed lags for ΔLog P can be expected to decline monotonously (in absolute terms) to zero. For estimation purposes, a dummy variable is included (Loot) that takes the value of 1 during the looting episodes of 1991 and 1992-93 (October-December 1991, and December 1992-February 1993). This dummy provides a limited correction to the aggregate index of economic activity (which, as indicated above, is a geometric interpolation of real GDP). In addition, the regression includes seasonal dummies, as government revenue is normally subject to marked seasonal fluctuations. The regression results are shown below (** and * indicate significance at the 1 percent and 5 percent levels, respectively):

The goodness of fit is excellent, but the coefficients for the first two lags of ΔLog P are not significant at the 1 percent level (the inclusion of Log P or ΔLog P with a three-month lag yields a coefficient and a t-value close to zero). While all coefficients have the expected signs, the slope of the distributed lags for ΔLog P is inverted relative to what could be expected—increasing, rather than decreasing, in absolute value. This could be due to the fact that ΔLog P is somewhat autocorrelated, a characteristic that may also prevent the detection of longer time lags. For the sake of parsimony, it is preferable to drop ΔLog P_t-1 to ensure that all coefficients are significant at the 1 percent level.

The goodness of fit has marginally diminished, but remains quite good (see Figure 2); standard mis-specification tests on the residuals are all satisfactory.¹⁸ The sum of the distributed lags remains about the same, at −1.75, compared with −1.77 in Regression 1. This indicates that the Tanzi effect has been quite powerful in Zaïre: with monthly inflation at 1 percent, 10 percent, or 20 percent a month, real government nonmining revenue drops by 1.8 percent, 16 percent, or 28 percent, respectively (in the arithmetic formulation, this would stem from an average collection lag of about 1½ months).

Regression 1.

Estimates of Real Government Revenue

(Monthly data, June 1990-June 1996 —seasonal coefficients not reported)

Regression 1.

Estimates of Real Government Revenue

(Monthly data, June 1990-June 1996 —seasonal coefficients not reported)

Regression on Log ROt/Pt	Coefficient	Standard error	t-value	Partial R2
Δ Log Pt	−0.4435	0.1909	−2.32 *	0.0879
Δ Log Pt-1	−0.5517	0.2289	−2.41 *	0.0940
ΔLog Pt-2	−0.7764	0.1892	−4.10 **	0.2311
Log yt-1	1.5264	0.2075	7.36 **	0.4914
Loot	−0.6905	0.1340	−5.15 **	0.3216
Constant	−4.0496	0.9037	−4.48 **	0.2639
	R2 = 0.8098	SEE = 0.2902		DW (1) = 1.85
	Adj. R2 = 0.7556	F (16,56) = 14.91 **		DW (12) = 1.84

View Table

Regression 1.

Estimates of Real Government Revenue

(Monthly data, June 1990-June 1996 —seasonal coefficients not reported)

Regression on Log ROt/Pt	Coefficient	Standard error	t-value	Partial R2
Δ Log Pt	−0.4435	0.1909	−2.32 *	0.0879
Δ Log Pt-1	−0.5517	0.2289	−2.41 *	0.0940
ΔLog Pt-2	−0.7764	0.1892	−4.10 **	0.2311
Log yt-1	1.5264	0.2075	7.36 **	0.4914
Loot	−0.6905	0.1340	−5.15 **	0.3216
Constant	−4.0496	0.9037	−4.48 **	0.2639
	R2 = 0.8098	SEE = 0.2902		DW (1) = 1.85
	Adj. R2 = 0.7556	F (16,56) = 14.91 **		DW (12) = 1.84

Regression 2.

Estimates of Real Government Revenue

(Monthly data, June 1990-June 1996 —seasonal coefficients not reported)

Regression 2.

Estimates of Real Government Revenue

(Monthly data, June 1990-June 1996 —seasonal coefficients not reported)

Regression on Log ROt/Pt	Coefficient	Standard error	t-value	Partial R2
ΔLog Pt	−0.7105	0.1619	−4.39 **	0.2525
ΔLog Pt-2	−1.0372	0.1617	−6.42 **	0.4193
Log yt-1	1.5211	0.2161	7.04 **	0.4650
Loot	−0.6639	0.1391	−4.77 **	0.2856
Constant	−4.0113	0.9409	−4.26 **	0.2418
	R2 = 0.7902	SEE = 0.3022		DW (1) = 1.91
	Adj. R2 = 0.7350	F (15,57) = 14.31 **		DW (12) = 1.76

View Table

Regression 2.

Estimates of Real Government Revenue

(Monthly data, June 1990-June 1996 —seasonal coefficients not reported)

Regression on Log ROt/Pt	Coefficient	Standard error	t-value	Partial R2
ΔLog Pt	−0.7105	0.1619	−4.39 **	0.2525
ΔLog Pt-2	−1.0372	0.1617	−6.42 **	0.4193
Log yt-1	1.5211	0.2161	7.04 **	0.4650
Loot	−0.6639	0.1391	−4.77 **	0.2856
Constant	−4.0113	0.9409	−4.26 **	0.2418
	R2 = 0.7902	SEE = 0.3022		DW (1) = 1.91
	Adj. R2 = 0.7350	F (15,57) = 14.31 **		DW (12) = 1.76

The large coefficient for real activity, in excess of 1.5, is at first surprising. However, it must be recalled that Log y does not reflect short-term fluctuations in output, some of which are captured in the dummy Loot. Overall, the sum of the coefficients for Log y and Loot (0.86) suggests that the Zaïrian tax system has had a rather low elasticity with respect to real economic activity.

B. The Price Equation

As a preliminary step, the structural equations (10) and (12) are estimated directly by imposing nonlinear restrictions on the coefficients. The results are summarized below:

Clearly, neither regression is acceptable. In both cases, a large part of the variance is captured by the constant term, and the correlation coefficients are rather small. In the RE equation, all the coefficients are significant at the 1 percent level, but α has the wrong sign, while β seems too large. All the coefficients for the AE equation have the expected positive sign, but α is excessively large, and β is not significantly different from zero. An examination of alternative estimates of these coefficients provides indications on the reasons for the instability of the regression. Figure 4 shows the recursive estimates of α, β, and λ in Regression 4 over successive 25-month periods (dates are centered; thus, for example, the estimates shown for January 1992 were made over the period July 1991-July 1992). It is noteworthy that the estimates from early 1993 to mid-1994—covering mid-1992 to late 1994—are relatively stable, especially considering that this period included the record hyperinflation episode of late 1993. By contrast, major deviations occur over the second half of 1992 and in late 1994/early 1995. These correspond to two periods of temporary disinflation, which suggest that actual inflationary expectations in Zaïre have been much more complex than assumed earlier (equation 7 above). In particular, this may indicate that expectations are asymmetric, or that they take into consideration significantly longer lags than specified above.

Recursive Estimates of Coefficients

(June 1990-July 1992 to May 1994-June 1996) ^1/

Citation: IMF Working Papers 1997, 050; 10.5089/9781451846935.001.A001

^1/ Coefficient and 5 percent confidence intervals (coefficient +/− two standard deviations). Regression estimates over 25-month periods, with dates centered.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Recursive Estimates of Coefficients

(June 1990-July 1992 to May 1994-June 1996) ^1/

Citation: IMF Working Papers 1997, 050; 10.5089/9781451846935.001.A001

^1/ Coefficient and 5 percent confidence intervals (coefficient +/− two standard deviations). Regression estimates over 25-month periods, with dates centered.

• Download Figure
• Download figure as PowerPoint slide

Recursive Estimates of Coefficients

(June 1990-July 1992 to May 1994-June 1996) ^1/

Citation: IMF Working Papers 1997, 050; 10.5089/9781451846935.001.A001

^1/ Coefficient and 5 percent confidence intervals (coefficient +/− two standard deviations). Regression estimates over 25-month periods, with dates centered.

• Download Figure
• Download figure as PowerPoint slide

Regressions 3 and 4.

Estimates of ΔLog P

(Monthly data, June 1990-June 1996)

Regressions 3 and 4.

Estimates of ΔLog P

(Monthly data, June 1990-June 1996)

	Regression 3			Regression 4
	(Equation 10 — RE)			(Equation 12 — AE)
	α	β	Const.	α	β	λ	Const.
Coefficient	−2.674	0.611	2.153	2.660	0.016	0.112	4.779
t-value	4.304	4.040	5.026	2.279	0.348	2.552	14.397
	R2 = 0.2704				R2 = 0.5636
	F (2,70) = 12.97				F (3,69) = 29.71

View Table

Regressions 3 and 4.

Estimates of ΔLog P

(Monthly data, June 1990-June 1996)

	Regression 3			Regression 4
	(Equation 10 — RE)			(Equation 12 — AE)
	α	β	Const.	α	β	λ	Const.
Coefficient	−2.674	0.611	2.153	2.660	0.016	0.112	4.779
t-value	4.304	4.040	5.026	2.279	0.348	2.552	14.397
	R2 = 0.2704				R2 = 0.5636
	F (2,70) = 12.97				F (3,69) = 29.71

To arrive at a better estimate of the price equation, an additional variable is included to reflect the six-monthly inflation rate, the restrictions on the coefficients in equation 12 are lifted, and seasonal dummies added. To improve the significance of the estimates, Log M_t-1 and ΔLog M_t are collapsed into Log M_t, while nonmining government revenue is dropped, considering that it duplicates the distributed lags of Log P and/or ΔLog P. The results are shown below:

All coefficients are significant at the 1 percent level, with particularly high t-values for Log P_t-1 and Log M_t, and the signs are those that would be expected. The coefficient for Log P_t-1, in particular, may be seen as a proxy for -λ in equation 12 above, although the changes made on the functional form prevent proper identification. While all standard mis-specification tests are satisfactory, a careful examination of the residuals shows systematic errors during 1994-95. A possible explanation is that the actual stock of currency was higher than reported, reflecting the parallel issuance of currency at the time when banknote printing was temporarily shifted to a less reputable company. To test this hypothesis, a dummy variable (D9495) is added, which takes the value of 1 from June 1994 through August 1995, and zero otherwise. The revised estimates are shown below:

While not very large in absolute terms, the dummy coefficient is significant at the 1 percent level, and the goodness of fit has improved noticeably. Moreover, all t-statistics are somewhat larger than in Regression 5. The estimates above would thus seem to confirm that there were substantial amounts of currency issued without proper accounting during this period, even though it is probable that the dummy also captured other special factors that were at play during the 1994-95 period.

Regression 5.

Estimates of Monthly Inflation

(Monthly data, June 1990-June 1996 —seasonal coefficients not reported)

Regression 5.

Estimates of Monthly Inflation

(Monthly data, June 1990-June 1996 —seasonal coefficients not reported)

Regression on ΔLog Pt	Coefficient	Standard error	t-value	Partial R2
Log Pt-1	−0.6705	0.0654	−10.26 **	0.6485
ΔLog Pt-1	0.3406	0.0799	4.26 **	0.2415
Log Pt-1/Pt-7	0.0908	0.0213	4.27 **	0.2423
Log Mt	0.7112	0.0695	10.24 **	0.6477
Constant	−3.4472	0.3583	−9.62 **	0.6189
	R2 = 0.8169	SEE = 0.1220		DW(l) = 2.1l
	Adj. R2 = 0.7687	F(15,57) = 16.95 **		DW(12) = 2.20

View Table

Regression 5.

Estimates of Monthly Inflation

(Monthly data, June 1990-June 1996 —seasonal coefficients not reported)

Regression on ΔLog Pt	Coefficient	Standard error	t-value	Partial R2
Log Pt-1	−0.6705	0.0654	−10.26 **	0.6485
ΔLog Pt-1	0.3406	0.0799	4.26 **	0.2415
Log Pt-1/Pt-7	0.0908	0.0213	4.27 **	0.2423
Log Mt	0.7112	0.0695	10.24 **	0.6477
Constant	−3.4472	0.3583	−9.62 **	0.6189
	R2 = 0.8169	SEE = 0.1220		DW(l) = 2.1l
	Adj. R2 = 0.7687	F(15,57) = 16.95 **		DW(12) = 2.20

Given these estimates, the long-term stationary solution to the price equation is:

However, the short-run elasticity of prices relative to the money stock is much higher than 1 because of the Tanzi effect, since the increase in prices depresses government revenue, which generates a deficit in government operations and thus a further increase in the money stock. The simulations below indicate that the short-term elasticity of prices relative to the money stock is about 1.25, while the long-run elasticity is on the order of 1.07.