Exchange rate management strategies in the accession countries: the case of Hungary.

Jones, Garett ; Kutan, Ali M.

INTRODUCTION

Hungary is the lead candidate to join the European Monetary Union (EMU) in the near future. EMU candidate countries must adhere to the new exchange mechanism, ERM II, and maintain parity between their currency and the euro within a [+ or -] 15% margin. This margin needs to be maintained for at least 2 years before they can qualify to join the euro area (Stage III of EMU). Along with this requirement, the candidate countries have to follow credible monetary and fiscal policies toward EMU reference values for inflation and interest rates, and government deficit and debt, respectively.

Leading candidate countries like Hungary face a severe conflict. Although stable exchange rates may bring about more rapid integration into the EU and relatively rapid nominal convergence, stable rates are likely to delay the catch-up process with the EU member countries in terms of productivity, income, wages, and wealth. The catch-up process implies that productivity, income, and wages levels in candidate countries must rise faster than those in member countries. Exchange rates in these countries must adjust to reflect these developments. If exchange rates cannot adjust, this will put pressure on the balance of payments and the real exchange rate, thus delaying the catch-up process (van Brabant, 2002).

Thus candidate countries need to choose an optimal exchange rate management policy during the interim period. In this paper, we focus on 'monetary policy' convergence issues. These issues should be relevant to policy-makers who are attempting to choose the optimal interim exchange rate policy. (1) More specifically, we measure the effect of shifts in Euro-area monetary policy during the 1990s, proxied by shifts in German interest rates and euro-area interest rates, on Hungarian industrial production and prices. (2) Although the discussion focuses on Hungary, the main points are likely to be applicable to other advanced candidate countries.

Our basic thesis is that if Hungarian and Euro-area output and prices respond in a similar way to shifts in Euro-area monetary policy, we would conclude that an early move to a fixed exchange rate with the euro would be feasible and desirable. If the Hungarian economy appears to be hypersensitive to shifts in Euro-area monetary policy, then a more flexible exchange rate regime would help insulate the Hungarian economy from the effects of Eurozone monetary policy. If the Hungarian economy appears to be relatively insensitive to shifts in Euro-zone monetary policy, then Hungary has a long way to go before it will be ready to join the Stage II of the EMU, for the process of joining the Euro will require a lengthy transition period during which macroeconomic institutions such as wage- and price-setting structures and trade and capital flows adjust to a new regime of Euro dominance (Kutan and Brada, 2000). A move toward a harder peg would be the recommended policy in this case.

The paper proceeds as follows: first we provide an overview of exchange rate policy in Hungary since the beginning of economic reforms in 1990. Next we offer historical and theoretical background for the exchange rate policy choices facing Hungary. Then we briefly describe our methodology and discuss our results along with the policy implications. The last section concludes the paper.

AN OVERVIEW OF EXCHANGE RATE POLICY IN HUNGARY (3)

At the beginning of the 1990s, monetary policy in Hungary included active exchange rate management based on a currency peg within a narrow band. The goal was real appreciation of the forint to help combat domestic inflation. This policy proved too costly because of declining competitiveness of Hungarian exports and sluggish growth. In addition, the policy failed to provide a nominal exchange rate anchor to reduce inflationary expectations. These costs began to appear in 1993, when the current account deficit reached 9% of GDP and then increased to 9.4% the following year. At the same time, the government's budget deficit remained unacceptably high (Table 1). The fact that the foreign debt was also growing steadily put Hungary at risk of insolvency. This macroeconomic situation was not sustainable.

During the 1990-1994 period, loose fiscal policy led to growing budget deficits and a high level of foreign debt. Financing this deficit required monetary expansion as well as high interest rates so that commercial banks would find government securities attractive. This policy fueled the inflation that the strong forint policy had sought to reduce. Continuous real appreciation led to loss of competitiveness and high current account deficits. The persistence of these twin deficits (Table 1) created uncertainty among Hungary's foreign creditors as well as concerns about the stability of the forint.

The two conflicting priorities of the government, controlling inflation and improving international competitiveness, led to speculation against the forint, which undermined the credibility of the exchange rate regime. Liberalisation of foreign exchange operations and the continuous real appreciation of the forint, resulted in significant capital inflows, gradually narrowing the ability of the monetary authorities to control the money supply. Moreover, during this period, there was no coordination between monetary and fiscal policy (Nemenyi, 1997).

The 1994 Mexican crisis further worsened Hungary's ability to borrow in international markets as the risk premium increased on emerging market debt. The government realised that it could not sustain the dual objectives of controlling inflation and improving international competitiveness at the same time and announced a major fiscal adjustment programme in March 1995 (Szapary and Jakab, 1998). Fiscal policy was tightened to reduce the twin deficits through lower government expenditures, higher import tariffs, and reduced government borrowing. Price stability was declared the key long-run goal of monetary policy.

The March 1995 measures included a major change in the nominal exchange rate regime. The change was intended to create credibility for economic policy, reduce the uncertainty associated with future policy measures, and restore inventors' confidence in the system (Nemenyi, 1997). Following a 9% devaluation of the official exchange rate, a preannounced crawling band exchange rate system was introduced. The crawl was based on a currency basket that consisted of the DM and U.S dollar with shares of 70 and 30%, respectively. In January 1998, the euro replaced the share of the DM. The band of permitted fluctuations was 2.25% on either side of the parity. This band was maintained until May 2001, when the band was widened to [+ or -] 15%. The rate of crawl was set according to an inflation target. The initial monthly rate was 1.9%. This was gradually reduced until it was 0.3 % in April 2000. In January 2000, the exchange rate and the basket were completely tied to the euro. Crawling devaluation created inflationary pressures and the inflation rate exceeded Poland's during this period. In addition, the National Bank of Hungary faced growing problems related to large capital inflows (Orlowski, 2001). In the face of these increasing problems, the Bank announced a policy of inflation targeting as of June 2000. The Bank publishes its inflation forecasts in its 'Quarterly Report on Inflation' to make the monetary policy more transparent. Year-on-year CPI inflation declined steadily from about 10% in 2001 to 5.9% in March 2002 (Hungary, IMF Country Report, 2002). However, this lower inflation also reflected declining food and fuel prices, besides the policy stance. The current inflation target is 3.5% for December 2003 and 2004 with a 1% tolerance band. However, according to the latest (November 2003) Quarterly Report on Inflation, inflation forecasts for this and next year are much higher than the 3.5% target, suggesting that inflation targeting policy has not been successful so far.

To summarise, the post-1995 exchange rate regime in Hungary focused on Ca) the stability of the nominal exchange rate as a tool of disinflation and (b) preventing significant real appreciation of the forint in order to sustain the current account balance and to control capital inflows (Orlowski, 2001). In addition, Hungary experienced some shock therapy, for example, the bankruptcy legislation introduced in the early 1990s. It received a significant amount of foreign direct investment. As a result, supply shocks should play a key role in output and price movements. The 1995 Bokros package also contained elements of shock therapy, including drastic cuts in budget spending, devaluation of the forint, change to a crawling peg, and the introduction of import surcharges. Finally, the National Bank of Hungary has moved toward greater exchange rate stability after May 2001 in the form of a wider exchange rate band. The ability to sustain the wide band will depend upon on the success of inflation targeting regime and policies that are consistent with the band itself.

HISTORICAL AND THEORETICAL BACKGROUND

In the aftermath of the Mexican and Asian financial crises of the 1990s, the range of policy choices has, in the words of Fischer (2001), been 'hollowed out' to two real alternatives: a hard peg (which would include dollarisation, a currency board, or a similar form of allegedly 'permanent' currency price-fixing) or a float (whether managed or unmanaged). We consider the two relevant policy choices for Hungary to be either a float or a hard peg to the Euro. As we will discuss below, pegging to the U.S. dollar would probably destabilise the economy.

As noted above, any discussion of interim policy must begin with Mundell's (1961) classic treatment of optimal currency areas; in particular, a key question inspired by this line of work is whether the accession countries experienced the same kinds of shocks as the Euro-area countries. To the extent that they experience similar shocks, it is better for the accession countries to peg to the Euro, since EMU policies will be responding to the same economic pressures as those affecting the accession countries. For example, if oil price shocks have similar effects on both the Euro-area and on Hungary, then the EMU will respond to an oil price shock in much the same way that Hungary would on its own; therefore, it will be low cost for Hungary to peg to the Euro.

The particular shocks we focus on are the monetary policy shocks produced by the European Central Bank itself. In a long line of work beginning with Christiano and Eichenbaum (1992) and summarised in Christiano et al. (2000), empirical macroeconomists have discovered that innovations to monetary policy as measured by impulse responses to vector autoregressions (VARs) can provide useful, robust information about how economies react to exogenous shifts in interest rates. In the thoroughly studied U.S. case, it is clear that unexpected increases in the federal funds rate cause a decline in output with a lag of about 6 months to a year, and to a decline in the price level within about 2 years. In the long run, these shocks appear to have no effect on output. As many have noted, this literature has provided substantial econometric support for the conventional view of monetary transmission as expressed in Friedman (1968).

We extend the techniques of the monetary policy shock literature to the question of optimal interim exchange rate policy in the accession countries. We use Euro-area monetary policy shocks to deduce whether Hungary would find it difficult to peg to the Euro at this time. We assume that if Hungary chooses a hard peg to the Euro, Hungary will then be at least as sensitive to Euro-area monetary policy shocks as the other EMU nations. This is because shifts in Euro short-term rates will be immediately transmitted to nations with hard Euro pegs. If instead Hungary chooses to float, Hungarian policy-makers will retain some monetary policy independence, despite the fact that Euro-area monetary policy shocks could have effects on the economy through shifts in imports, exports, or the exchange rate.

METHODOLOGICAL ISSUES

We are interested in the time-series properties of an ((m + n) x 1) vector of covariance-stationary (4) variables, Yr. We will stack the variable so that the m variables from the large country (denoted [y.sup.L.sub.t]) are on top, and the n variables from the small country (denoted [y.sup.S.sup.t]) are on the bottom. The m large country variables might contain this month's values of German industrial production and short-term German interest rates, for example. In such a case, m would equal two. The vector is denoted as follows:

(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

In the structural vector autoregression (SVAR) literature, the current values of [y.sub.t] depend partially on the lagged values of all of the elements of [y.sub.t]: For example, this month's unemployment rate might be related to last month's unemployment rate and last month's level of industrial production. The number of lags is denoted by p.

In addition, some elements of this month's [y.sub.t] will be affected by other elements of this month's [y.sub.t]. For example, when the Federal Reserve sets the federal funds rate, their decision will depend on that month's consumer price index and that month's industrial production. This latter situation--where some elements of [y.sub.t] depend on other elements of [y.sub.t], could conceivably lead to a classic problem of simultaneous equations bias. As we will see, however, econometric techniques have been developed (summarised by Christiano et al., 1999) that provide a tractable solution to this potential problem. Therefore, let us consider the following equation, which jointly characterises the large-country economy and the small-country economy we are interested in:

(2) [y.sub.t] = [alpha] + [[PHI].sub.0][y.sub.t] + [[PHI].sub.1][y.sub.t-1] + ... + [[PHI].sub.p][y.sub.t-p] + [[epsilon].sub.t]

Here, [alpha] is an (m + n) x 1 vector, the [PHI] matrices are (m + n) x (m + n), and [epsilon.sub.t] is a vector of (m + n) x 1 shocks to both the large economy and the small one. These shocks [epsilon.sub.t] are assumed to be independently and identically distributed, with a diagonal covariance matrix, denoted D. This implies that there are a total of (m + n) shocks hitting the economy each period, but while the diagonal covariance matrix implies that the same-period shocks are uncorrelated with each other, it is possible for each shock to influence more than one element of [y.sub.t].

For instance, if there is a shock to industrial production this month, that will, of course, immediately influence industrial production; and since this month's federal funds rate depends in part on this month's industrial production, the industrial production shock will simultaneously influence this month's federal funds rate. The combination of same-period relationships between the elements of [y.sub.t] is contained within the matrix [[PHI].sub.0]. There must be some restrictions on the coefficients of [[PHI].sub.0] in order to estimate its coefficients; the first restriction is that the diagonal elements must be zero. Other sufficient restrictions will be noted later.

We begin by assuming that large-country economic variables can have an effect on small-country economic variables, but that small-country economic variables have no effect, either in the current period or in the future, on the large-country economic variables. Another way of stating our assumption is that the small-country economic variables contain no unique information about the dynamic behaviour of the large-country economy. This assumption imposes the following restriction on the shape of the [[PHI].sub.i] matrices, for i = 0, 1 ..., p, where [[PHI].sup.X,Y.sub.i] is an appropriately shaped submatrix:

(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

Given this form, the small-country variables have no feedback effects on the large-country variables; combine this with the assumption that the [[epsilon].sub.t] shocks are uncorrelated with each other, and we can conclude that the small-country economy has no effect on the large-country economy. (5)

Next, assuming that [[PHI].sub.0] is invertible, we can make the following transformation:

(4) (I - [[PHI].sub.0])[y.sub.t] = [alpha] + [[PHI].sub.1][y.sub.t-1] + ... + [[PHI].sub.p][y.sub.t-p] + [[epsilon].sub.t],

(5) [y.sub.t] = [(I - [[PHI].sub.0]).sup.-1][alpha] + [(I - [[PHI].sub.0]).sup.-1][[PHI].sub.1][y.sub.t-1] + ... + [(I - [[PHI].sub.0]).sup.-1][[PHI].sub.p][y.sub.t-p] + [(I - [[PHI].sub.0]).sup.-1][[epsilon].sub.t]

and if we define a = [(I - [[PHI].sub.0]).sup.-1] [alpha], and [P.sub.i] = [(I - [[PHI].sub.0]).sup.-1][[PHI].sub.i], this can be rewritten as

(6) [y.sub.t] = a + [P.sub.1][y.sub.t-1] + ... + [P.sub.p][y.sub.t-p] + [(I - [[PHI].sub.0]).sup.-1][[epsilon].sub.t].

Note that in this format, the covariance matrix of [e.sub.t.] is now ([(I - [[PHI].sub.0]).sup.-1])D([(I - [[PHI].sub.0).sup.-1])', and hence is unlikely to be a diagonal matrix. This implies that the contemporaneous relationships between the elements of [y.sub.t] are now reflected in the covariance matrix of the disturbance terms.

Since the [[PHI].sub.i] matrices are block lower-triangular, this implies that the [P.sub.i] matrices and the [(I - [[PHI].sub.0]).sup.-1] matrix will all remain block lower-triangular; in words, this means that (6) represents a world where the small country cannot affect the large one. Therefore, we can estimate large-country coefficients ([[PHI].sub.0.sup.L,L]) without including small-country data in the regressions.

Since we are interested in the effects of large-country monetary policy shocks on a small accession country like Hungary, we want to create a time series of these shocks. This time series is one of the large-country elements of [[epsilon].sub.t]; we will denote it [[epsilon].sub.t.sup.Lm]. The simplest way to estimate [[epsilon].sub.t.sup.Lm], as in Rudebusch (1996), is to regress our monetary policy variable, the short-term interest rate, on its own lags, lags of all large-country variables, and current values of large-country variables that appear higher in the Choleski ordering (details are provided in the next section). Following standard practice in the monetary policy literature, we include 1 year of lags (p = 12 for monthly regressions), and denote the series of residuals [[epsilon].sub.t.sup.LM].

Note that we can now write the small country's time-series properties as follows:

(7) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where, as before, L denotes the large country, and S the small. Clearly, estimating equation (7) directly would require a large number of observations, since each [P.sub.i.sup.S,S] matrix has n x n parameters, and each [P.sub.i.sup.L,S] matrix has m x n parameters. Since the post-communist nations have, at best, 120 monthly observations, one would quickly run out of the bare minimum of (m + n) x n x p observations needed to estimate this equation (especially if p = 12 for monthly regressions). Fortunately, we can avoid the need to estimate the entire set of [P.sub.i.sup.L,S] coefficients if we take advantage of the fact that every stationary vector autoregression has an infinite-order moving average representation. Since the large country is unaffected by the small country, we focus on the vector moving average representation for the large country.

Hamilton (1992, pp. 257-260) demonstrates that any covariance-stationary vector process can be rewritten as an infinite-lag vector moving-average process. This is the vector extension of the fact that any finite-order autoregressive process has an infinite-order moving average representation.

Therefore, we have:

(8) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

If the process is covariance-stationary, then the infinite lags will approach zero matrices. Substituting lags of (8) into the [P.sub.i.sup.L,S][y.sup.L.sub.t-1] + ... + [P.sup.L,S.sub.p][y.sup.L.sub.t-p] terms from (7), we can eliminate all lagged [y.sup.L] values from (7), replacing them instead with a cumbersome set of p infinite moving-average processes. We collect the moving-average coefficient matrices for each of the large-country shock terms together, and denote the matrix for the [[epsilon].sup.L.sub.t-1] shock by [[lambda].sub.i].

Note that this elimination of lagged [y.sup.L] values will also impact the constant. We denote the constant of this new process as [a.sup.s], which is equal to [a.sup.s] (the constant from the VAR representation) plus [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. By simplifying the constant and vector moving-average coefficients in these ways, we have:

(9) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

At this point, we see that the small-country economy follows a time-series process superficially similar to an ARMA process; and in order to achieve unbiased estimates of the coefficients, we would need to include all the elements of [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], that have non-zero coefficients in the [[lambda].sub.i] matrix. In this paper, we employ the working assumption that large-country monetary policy shocks are the only large-country shocks affecting the small-country economy. This assumption is the same as assuming that only the column of [[lambda].sub.i] (i = 0, 1, ..., p) corresponding to [[epsilon].sup.Lm] has non-zero elements.

This assumption is not as strong as it may appear: in claiming that monetary policy shocks are the only large-country shocks that matter, we are not claiming that other elements of the [y.sup.L] have no effect on the small country; indeed, the moving-average [[PSI].sub.i] matrices are complicated functions of the elements of the large country's VAR coefficients. So, if a large-country monetary policy shock slows large-country output 6 months later, this is captured in a [[PSI].sub.6] coefficient, and hence is captured in a [[PSI].sub.6] coefficient. Accordingly, large-country monetary policy can have an effect on small-country economic variables through more than one channel.

Since our goal is to estimate impulse-response functions that demonstrate the effect of large-country monetary policy shocks on a small country, we will estimate (9) by including the monetary policy shocks as an additional variable in the small-country VAR. This method has the benefit of being tractable, although it implies inefficiently large standard errors, since we do not impose zero restrictions on the monetary policy shock lag coefficients. In brief, we can think of our econometric model as a small-country VAR with the added exogenous moving-average shock of large-country interest rate surprises. It is also important to bear in mind that such identified monetary shocks may also include anticipated future inflation shocks, including commodity price movements such as oil and food, in addition to unexpected movements in monetary policy. Our results regarding the effect of EMU entry on Hungary's business cycle can therefore not be considered definitive, and should be interpreted accordingly.

DATA AND OVERVIEW OF EMPIRICAL METHODOLOGY

Data came from the IMF's International Financial Statistics database, the OECD's Main Economic Indicators database, and the European Central Bank's time series database. The data used are all monthly, and include seasonally adjusted industrial production, the level of consumer prices, the trade-weighted exchange rate (nominal in the case of Germany due to a longer dataset, real for all other countries), and a short-term nominal interest rate.

In the first set of regressions, the sample period is 1992:1-1998:12 for Hungary, 1947:1-2001:6 for the United States, and 1960:1-1998:12 for Germany. We use the behaviour of the German economic variables as a proxy for the future behaviour of the ECB. We end this sample in 1998 because of the regime switch that occurred when the European Central Bank took over monetary policy for Germany and the other Euro-area countries, when German interest rate data became no longer available.

In the second set of regressions, we use a monthly sample period from 1994:1 to 2003:7. In this second set of regressions, Euro-area data are available, including, most importantly, the interbank lending rate among Euro-area countries. A Euro-area industrial production index, a real exchange rate, and a harmonised price index are also available. This short sample, which overlaps the first sample, provides a much-needed robustness check on the earlier results.

German and U.S. monetary policy shocks were estimated in the standard way: for the U.S., log industrial production (IP), log consumer prices (P), log producer prices (PPI), the federal funds rate (FFR), the log of non-borrowed reserves (NBR), and the log of total reserves (TR) were included in a VAR with 12-monthly lags, their ordering in the Choleski decomposition was IP, P, PPI, FFR, NBR, TR. Here, as is typical in the monetary policy shock literature, our 'shocks' are innovations to FFR that are uncorrelated with 12 lags of the variables in the information set, as well uncorrelated with current values of variables that come earlier in the Choleski ordering.

As Rudebusch (1996) and Christiano et al. (1999) note, these FFR monetary policy shocks are equivalent to the residuals from a regression of the federal funds rate on current values of IP, P, and PPI, as well as on 12-monthly lags of IP, P, PPI, FFR, NBR, and TR. Accordingly we run such a regression, and label the residuals from this regression as Federal Funds rate shocks (FFS).

As proxies for German monetary policy shocks, we follow a similar procedure, including German values of IP, P, the nominal exchange rate (NER), and the rate on overnight call money (CALL) in a 12-monthly lag VAR, using the Cholesky ordering IP, P, NER, CALL. The shocks themselves are the residuals from a regression of CALL on current values of IP, P, and NER, as well as 12-monthly lags of IP, P, NER, and CALL. Impulse-response functions showing the estimated effects on German economy of own-country monetary policy shocks are shown in Chart 1, and the German variance decomposition is reported in Table 2. The effects of monetary policy shocks on output accord with the conventional wisdom: monetary tightenings lead to persistent declines in output a few months later. The effects on prices are anomalous, but this is likely the result of the well-known 'price puzzle,' a result of the fact that central banks tighten interest rates when they see inflation 'in the pipeline' by referring to producer price data not included in the VAR. Since such producer price variables are omitted for both German and accession country VARs, our price results should be comparable between these two groups.

For Hungary, we ran VARs that included the German monetary policy shocks, log of industrial production, log of consumer prices, log real exchange rate, and a private-sector short-term interest rate. We ran the VARs in levels and differences, and in 6- and 12-month versions, with no important difference in the results. We focus our discussion entirely on the 6-month VAR in levels because the 12-month version had the primary effect of increasing the standard errors.

We did not include the US and German shocks together because they are not positively correlated with each other (correlation coefficient = -0.08) and because we would lose additional degrees of freedom. The US monetary policy shocks turn out to be of no interest since they explain almost none of the variance in output or prices in Hungary. Overall, the pattern is clear: Germany, not the US, is the economically important source of monetary policy shocks for Hungary. Therefore, we will turn our focus to the effect of German monetary policy shocks, and using the second dataset, Euro-area monetary policy shocks.

RESULTS AND DISCUSSION

The first fact to notice is that output in Hungary is far more sensitive than output in Germany to German monetary policy shocks (Charts 1 and 2). Industrial production responds twice as strongly in Hungary compared to the German industrial production response.

[ILLUSTRATIONS OMITTED]

Another important fact revealed by impulse responses is that the fall in industrial production occurs much earlier in Hungary than it does in Germany; Germany's response is negligible 6 months after the shock, and slowly moves toward its low point after two full years. By contrast, Hungary reaches its low points well within a year, and recovery appears to be over by the time the Germans are reaching their low point. Owing to the asynchronous response of output to German monetary policy shocks, we can be sure that shifts in Euro-area interest rates will be a cause for concern in Hungary well before they register in the Euro-area macroeconomy: this provides an additional reason for Hungary to be cautious about tying their exchange rates more closely to the Euro-area.

Notably, Hungary shows much more modest output responses to their own country's interest rate shocks (Chart 3). This could reflect the thinner domestic financial markets and indicate that the within-country monetary transmission mechanism works much differently than the between-country monetary transmission mechanism already documented. (6) One can think of this as an Investment-Saving (IS) curve framework, where the output in Hungary depends on the real German interest rate as well as the real interest rate in Hungary: these results indicate that the coefficient on the German interest rate would be about an order of magnitude larger than the coefficient on the interest rate in Hungary.

[ILLUSTRATION OMITTED]

Variance decompositions for output bolster the results from impulse-response functions (Table 3). German monetary policy shocks explain about 20% of the variance of output in Hungary over a 4-year horizon, while domestic interest rate shocks almost always explain less than 13% of output volatility over the same horizon. Clearly, IS curve in Hungary is much more sensitive to German interest rates (and by extension, Euro-area interest rates) than they are to their own interest rates.

Turning to results regarding the price level, the variance decompositions in Charts 4 and 5 indicate that German monetary policy shocks explain little of the volatility in prices over the period. We should conclude, therefore, that one-time shifts in Euro-area interest rates are unlikely to be an important driving force behind inflation or deflation in Hungary should they choose to fix their rates to the Euro.

ROBUSTNESS CHECK

Considering the short datasets available when one studies accession economies, any possible robustness check is valuable. The robustness check we consider is a set of 2-month and 6-month vector autoregressions using all available Euro-area data since the publication of Euro-priced interbank lending rates in 1994. This allows us to run 9-year VARs that roughly reinforce--and certainly do not contradict--the conclusions derived from the Germany-as-proxy VARs. The 2- and 6-month lag structures were chosen because traditional VAR lag length tests recommended lags in the 2-3 month range, as well as lags in the 6-month or longer range. We judged that with the short dataset available, moving beyond 6 months would have cost too many degrees of freedom, and would have resulted in overfitting the data. Accordingly, using the same techniques as in the German case, we estimated 2- and 6-month Euro-area monetary policy shocks, and saved these residual shocks as separate variables. We then used the 2-month monetary policy shocks in a 2-month Hungarian VAR, and used the 6-month monetary policy shocks in a 6-month Hungarian VAR, each estimated as before. Producer price indices were omitted both to save degrees of freedom, and because they did not appear to solve any 'price puzzle' problems.

[ILLUSTRATIONS OMITTED]

When we consider the second set of vector autoregressions based on Euro-area monetary policy shocks from 1994 to 2003, we see that the earlier results are reinforced. For brevity, we focus on the response of the Euro-area economy and the Hungarian economy to Euro-area interest rate shocks, and only report results for Hungarian industrial production. As before, if we find that the Hungarian economy is much more sensitive to European monetary policy shocks than the Euro-area itself is, we interpret such excess sensitivity as a reason to prefer a flexible exchange rate in the lead-up to joining the Euro.

Charts 6 and 7 demonstrate that Euro-area monetary policy shocks impact European industrial production in much the same way that German shocks impacted German industrial production: In each case, industrial production falls by about 1% within 2 years of a 1% monetary policy shock. Therefore, our assumption that Germany would be a good proxy for the Euroarea appears to be a plausible one. Further, the prize puzzle does not appear in the 6-month VAR, and is weak in the 2-month VAR: monetary tightenings most likely lead to lower prices after two and a half years under either specification. These price level results are not significant at the 95% confidence level, however.

[ILLUSTRATIONS OMITTED]

The response of Hungarian industrial production to these Euro-area shocks (Chart 8) appears much the same as before in the 2-month VAR: production appears to drop by about 2 % over the course of the first year after a shock, responding about twice as much and about twice as rapidly as the Euro-area economy to the same 1% interest rate shock. The response is between one and two standard deviations away from zero, so while we cannot be absolutely confident of the results, this is useful as a robustness check on the German results. The 6-month VAR is noisier, and the standard deviations are much wider, so little can be concluded from the 6-month result, either positively or negatively.

[ILLUSTRATION OMITTED]

Overall, the 1994-2003 data are quite consistent with the view that German monetary policy shocks are a good proxy for Euro-area monetary policy shocks, and give some support to our earlier conclusion that Hungarian industrial production responds at least twice as quickly and twice as dramatically to Euro-area monetary policy shocks, when compared against the Euro-area's own response.

CONCLUSIONS AND IMPLICATIONS FOR EXCHANGE RATE MANAGEMENT

For the accession countries, choosing a pre-Euro exchange rate regime is a difficult policy question: researchers only have a few years of data to work with, and the inherent complexity of small, open economies increases the number of variables that are relevant for our analysis. Econometric studies with few observations and many relevant variables present understandable difficulties for researchers. In this paper, we modified the canonical vector autoregression (VAR) model in order to deal with these difficulties.

Based on the vector autoregression results discussed here, it appears that industrial production in Hungary is extraordinarily sensitive to shifts in German interest rates. To the extent that German monetary policy is a good proxy for future Euro-area monetary policy, Hungary should be concerned about the increased output volatility that could follow from a policy decision to fix their exchange rate to the Euro in the near future, since a rise in German interest rates has an effect on Hungarian output that is likely at least twice as large as the effect on German output, and the effect on Hungary probably lasts for over a year. The results from the past decade, using Euro-area data, reinforced these conclusions, and demonstrated that Germany was indeed a good proxy for the Euro-area.

These results point out an overlooked cost to pegging the forint: the decision to peg means a decision to be buffeted by the Euro-area's monetary policy shocks. While weighing the overall costs and benefits of pegging is beyond the scope of this paper, these results certainly weigh heavily on the cost side of the ledger. If, instead of pegging, Hungary chose to follow an inflation-targeting framework, whether loose or strict, the nation would retain enough policy independence to insulate itself from the large effects of Euroarea monetary policy shocks. Under inflation targeting, Hungary would have the freedom to lean against this off-ignored European wind.

Acknowledgements

The author thank Balazs Vonnak of the National Bank of Hungary for very useful comments and Haigang Zhou for helpful research assistance. This paper was written while Garett Jones was an economic advisor at the US Senate. The usual disclaimer applies.

Table 1: Macroeconomic indicators; 1991-2001

 1991 1992 1993 1994 1995 1996

Real GDP Growth (%) -11.9 -3.1 -0.6 2.9 1.5 1.4
CPI inflation (%) 34.2 23.0 22.5 18.9 28.3 23.5
In % of GDP:
Current account balance 0.8 0.9 -9.0 -9.4 -5.6 -3.7
Government deficit (a) 3.0 7.0 6.5 8.4 6.7 3.1
Public debt
Consolidated (b) 66.9 65.0 84.3 83.2 85.4 71.7
Non-consolidated (c) 74.7 79.0 90.8 88.3 86.5 72.6

 1997 1998 1999 2000 2001

Real GDP Growth (%) 4.6 4.9 4.2 5.2 3.8
CPI inflation (%) 18.3 14.3 10.0 9.8 9.2
In % of GDP:
Current account balance -2.1 -4.8 -4.3 -2.8 -2.1
Government deficit (a) 4.9 4.8 3.7 3.7 3.0
Public debt
Consolidated (b) 63.6 63.5 66.6 62.1 57.1
Non-consolidated (c) 63.7 62.1 61.1 56.0 52.8

(a) Based on official data reported in Kiss and Szapary (Table 2,
2000), which include transactions of the central government, the
social security funds, the local authorities, and the
extra-budgetary funds. For details, see Kiss and Szapary (2000).

(b) Government and National Bank of Hungary, excluding the
sterilisation instruments of the central bank, see Kiss and
Szapary (Table 1, 2000).

(c) Non-consolidated with the National Bank of Hungary. For
details, see Kiss and Szapary (Table 1, 2000)

Table 2:

Month Log IP Log CPI Log ER Call rate

Variance Decomposition of German log IP

12 95.3 0.4 2.4 1.9
 (4.2) (1.5) (3.5) (1.8)
24 81.4 0.7 3.1 14.8
 (9.0) (2.6) (5.2) (7.6)
36 68.3 1.0 2.9 27.8
 (12.2) (3.7) (6.6) (12.1)
48 60.2 0.9 4.3 34.6
 (13.8) (4.1) (8.7) (14.6)

Variance decomposition of German log CPI

12 0.4 93.9 0.9 4.8
 (1.9) (4.7) (2.0) (4.0)
24 5.5 82.3 0.6 11.6
 (5.5) (9.3) (3.2) (7.7)
36 19.9 66.9 0.6 12.5
 (9.6) (11.5) (3.8) (9.2)
48 36.4 52.9 1.7 8.9
 (11.8) (11.9) (5.2) (8.3)

Standard errors are reported in parentheses.

Table 3:

Month GX Log IP Log CPI Log ER Call rate

Variance decomposition of Hungarian log IP

12 22.9 51.8 9.9 9.5 5.9
 (12.6) (10.6) (6.3) (7.1) (5.6)
24 24.6 43.7 13.4 10.4 7.9
 (13.8) (11.4) (7.7) (9.8) (10.9)
36 23.1 43.2 14.2 11.3 8.2
 (14.8) (13.0) (9.3) (12.5) (12.1)
48 20.2 38.2 17.9 11.0 12.6
 (15.6) (13.2) (10.3) (13.9) (12.4)

Variance Decomposition of Hungarian log CPI

12 4.0 2.0 61.0 0.6 32.5
 (9.0) (5.2) (13.0) (4.9) (11.1)
24 5.8 7.6 40.3 1.1 45.2
 (11.4) (8.6) (15.0) (7.3) (13.4)
36 9.7 16.9 26.3 1.0 46.1
 (13.7) (13.4) (16.3) (10.2) (15.3)
48 14.7 22.2 17.1 0.6 45.4
 (14.5) (15.8) (16.5) (10.8) (16.1)

GX is the estimate of German monetary policy shocks. Standard
errors are reported in parentheses.

(1) Our analysis is only partial because we discuss only monetary policy, For Hungary, other aspects, especially higher economic growth due to trade deepening or vulnerability of exchange rate regimes, also need to be taken into account.

(2) German interest rates are used because of the key role played by the German economy in the region, the evidence that other non-EMU members tend to tie their monetary policy to that of Bundesbank, and the general belief that the European Central Bank (ECB) will follow an anti-inflation policy as implemented by the Bundesbank in the past. For similar applications see, among others, Brada and Kutan (2001). Gottschalk and Moore (2001) also use Germany as a proxy for 'foreign shocks' to estimate the impact of external shocks on Polish prices.

(3) This section draws on Kutan and Brada (2000) and Dibooglu and Kutan (2001).

(4) In empirical work, the assumption of covariance-stationarity is generally violated, but this violation does not appear to affect the validity of the econometric results in any important way. Christiano et al. (2000) is a canonical example.

(5) Cushman and Zha (1997) impose the same zero restrictions in a small-country VAR, using the U.S. and Canada as the large and small countries, respectively. They do not impose a Choleski ordering on the [[PHI].sub.0] matrix, however, instead choosing their own set of cross-equation restrictions on [[PHI].sub.0].

(6) These results are supported by related studies. Using a VAR model, Gottschalk and Moore (2001) examine the significance of domestic interest rates and other variables on output movements in Poland during the 1992-99 period. They report that 'there is practically no role for the interest rate' (p. 35) to explain changes in output. However, this study does not account for the impact of foreign interest rates, as we do in this paper.

REFERENCES

Brada, JC and Kutan, AM. 2001: The convergence of monetary policy between candidate countries and the European Union. Economic Systems 25(3): 215-231.

Christiano, L and Eichenbaum, M. 1992: Identification and the liquidity effect of a monetary policy shock. In: Cukeriman, A, Hercowitz, Z and Leiderman, L (eds). Political Economy, Growth, and Business Cycles. MIT Press: London. pp. 335-370.

Christiano, L, Eichenbaum, M and Evans, CL. 1999: Monetary policy shocks: what have we learned and to what end? In: Taylor, JB and Woodford, M (eds). Handbook of Macroeconomics, Amsterdam: North-Holland. Vol. 1A, pp. 64-148.

Cushman, DO and Zha, T. 1997: Identifying monetary policy in a small open economy under flexible exchange rates. Journal of Monetary Economics 39: 453-448.

Dibooglu, S and Kutan, AM. 2001: Sources of real exchange rate fluctuations in transition economies: the case of Hungary and Poland. Journal of Comparative Economics 21(13): 1-19.

Fischer, S. 2001: Distinguished lecture on economics in government--is the bipolar view correct? Journal of Economic Perspectives 11 (Spring): 3-24.

Friedman, M. 1968: The role of monetary policy. American Economic Review 58: 1-17.

Gottschalk, J and Moore, D. 2001: Implementing inflation targeting regimes: the case of Poland. Journal of Comparative Economics 29: 24-39.

Hamilton, JD. 1992: Time series analysis. Princeton University Press: Princeton, NJ.

Kiss, GP and Szapary, G. 2000: Fiscal adjustment in the transition process: Hungary, 1990-1999. Post-Soviet Geography and Economics 41 (4): 233-264.

Kutan, AM and Brada, JC. 2000: The evolution of monetary policy in transition economies. Federal Reserve Bank of St Louis Review 82 (2): 31-40.

Mundell, R. 1961: A theory of optimum currency area. American Economic Review 50: 657-665.

Nemenyi, J. 1997: Monetary policy in Hungary: strategies, instruments and transmission mechanism. Focus on Transition, Oesterreichische Nationalbank 2: 131-161.

Orlowski, LT. 2001: From inflation targeting to the Euro-Peg: a model of monetary convergence for transition economies. Economic Systems 25(5): 235-251.

Rudebusch, G. 1996: Do measures of monetary policy in a VAR make sense? Working Papers in Applied Economic Theory 96-05. Federal Reserve Bank of San Francisco.

Szapary, G and Jakab, ZM. 1998: Exchange rate policy in transition economies: the case of Hungary. Journal of Comparative Economics 26(4): 691-717.

van Jozef Brabant. 2002: Exchange rate policy, EU integration, and catch-up modernization. Russian and East European Finance and Trade 38(1): 5-30.

GARETT JONES (1) & ALI M. KUTAN (1,2)

(1) Department of Economics and Finance, Southern Illinois University Edwardsville, Edwardsville, IL 62026, USA. E-mail: garjone@siue.edu, akutan@siue.edu;

(2) Center for European Integration Studies (ZEI), Bonn, Germany