Money Demand in Post-Crisis Russia: Dedollarization and Remonetization

Iikka Korhonen and Aaron Mehrotra

AbsTrAcT: This paper assesses the monetary determinants of inflation in Russia using money demand functions. We find a stable money demand relation for Russia following the 1998 crisis. Higher income boosts demand for real ruble balances and the income elasticity of money is larger than unity, reflecting remonetization in the Russian economy. Inflation affects the adjustment toward equilibrium, whereas broad money shocks lead to higher inflation. We also show that exchange rate fluctuations considerably influence Russian money demand. Our results for system stability and the predictive value of money justify using the money stock as an information variable. They also suggest that the strong influence of exchange rate on money demand is likely to continue, despite the dedollarization of the Russian economy. KEy words: dollarization, money demand, Russia, vector error-correction models.

The Russian economy has undergone substantial changes since the August 1998 economic and financial crises. Economic growth resumed fairly soon after the rubles drastic devaluation and a rebound in world energy prices. Rapid economic growth and soaring commodity prices have helped push the Russian governments finances into strong surplus after a prolonged period of public-sector deficits. The changes in the spheres of exchange rate and monetary policy, though less apparent, have also been profound. Before the August 1998 crisis, the ruble was pegged fairly tightly to the U.S. dollar. Since the crisis and the failure of the exchange rate pegthe linchpin of the Central Bank of Russias (CBRs) stabilization effortsthe CBR has avoided explicit exchange rate targets. It has, nevertheless, influenced changes in the rubles exchange rate quite heavily. At the same time, the CBR has had a target band for inflation. However, this target has usually been overshot, as the CBR has given precedence to maintaining the nominal exchange rate close to a desired level. In such an environment, searching for inflation determinants has obvious policy relevance. Our results should be interpreted in this light, as we assess how shocks to the broad money stock, as well as exchange rate changes affecting the demand for domestic money, are transmitted to inflation. Our results are novel to the literature on Russian monetary and exchange rate policy and, in addition, are relevant for other emerging market countries where the exchange rate continues to be used as a policy tool, even if a fixed exchange rate policy has been abandoned.

Iikka Korhonen (iikka.korhonen@bof.fi) is head of the Bank of Finland Institute for Economies in Transition (BOFIT). Aaron Mehrotra (aaron.mehrotra@bof.fi) is a senior economist at BOFIT. The authors thank Markus Lahtinen, Alexey Ponomarenko, and Jouko Rautava for their helpful comments on several versions of this paper. They are also grateful to the participants at the TwentyNinth Annual Meeting of the Finnish Society for Economic Research, Eighty-Second Annual Conference of the Western Economic Association, and internal seminars at the central banks of Russia and Finland. All opinions expressed here are those of the authors and do not represent the official position of the Bank of Finland.
Emerging Markets Finance & Trade / MarchApril 2010, Vol. 46, No. 2, pp. 519. Copyright 2010 M.E. Sharpe, Inc. All rights reserved. 1540-496X/2010 $9.50 + 0.00. DOI 10.2753/REE1540-496X460201

Emerging Markets Finance & Trade

In this paper, we estimate money demand functions for Russias postcrisis period. The evolution of the monetary policy regime over eight postcrisis years provides sufficient data to test econometrically for the presence of a stable money demand function. Several interesting research questionsall with clear policy implicationscan be assessed within this framework. First, we can assess whether a stable money demand function can be said to exist. We know that Russian households and companies have shown increasing trust in their currency and banks in recent years, which has led to a remonetization of the economy. But can we find a statistically significant and stable relation among money, prices, income, and other relevant variables in such an environment? If so, does, for example, the income elasticity of money demand differ from the values typically found in more mature economies? Second, we are interested in the role of the exchange rate in Russian money demand. Even though the demand for ruble deposits has clearly increased, foreign currency (especially U.S. dollar) holdings of Russian residents remain large. We want to know whether the exchange rate affects ruble money demand in Russia. Presumably, the larger the absolute value of this influence, the higher the remaining degree of dollarization. This would justify a large role for the exchange rate in conducting monetary policy, which is likely the case in other emerging market economies, where the exchange rate continues to be significant in setting monetary policy. To answer the above questions, we first look for a cointegrating relation among the real ruble money stock, an indicator for gross domestic product (GDP), and various proxies for the opportunity cost of holding money. We find that when these variables are augmented by a trend, a stable long-run relation among them seems to exist. This relation can be interpreted as a money demand function, whereas the trend variable captures the monetary deepening we see in postcrisis Russia. We also find that the data support including the exchange rate variable. Next, we employ a vector error-correction model to study the dynamics of our estimated system. A positive shock to real money stock increases inflation and output in the Russian economy. An inflation forecast based on the estimated model captures actual price dynamics reasonably well. These results suggest that money can usefully indicate future inflation pressures and that the exchange rate continues to influence money demand in Russia. The novelty of this analysis is that we concentrate solely on postcrisis data, which allow us to draw more relevant policy conclusions. The years before the 1998 crisis were marked by both a different monetary policy regime and the transition from a socialist system toward a more market-based system. The dynamics between the variables may not be regime independent, and estimating a money demand relation covering two different regimes could yield spurious results. Previous research has often found evidence of an unstable demand function for the ruble money stock (see, e.g., BahmaniOskooee and Barry 2000; Oomes and Ohnsorge 2005). Our investigation for model dynamics covers aspects that have not been investigated in money demand studies for Russia during the current decade. These include using the money demand system to forecast inflation, measuring uncertainty in impulse response dynamics, and valuing the information from the real ruble money gap, that is, the difference between actual and equilibrium ruble money stock. All of these are important in considering the use of money stock in the conduct of monetary policy. We also test for the importance of the spot and one-month forward ruble/U.S. dollar exchange rates as opportunity cost variables. This more clearly indicates the significance of exchange rate movements for ruble money demand.

MarchApril 2010 7

Description and Selective Literature Survey of Russian Monetary Policy During the Postcrisis Era The August 1998 financial crisis caused more than an abandonment of the fixed peg for the ruble; it forced the Russian government to reschedule its loan commitments, an act tantamount to partial default on its debts. Many commercial banks went under, some of them owing substantial sums of money to foreign investors who had hedged their ruble risks by selling ruble forward contracts. The underlying cause of the crisis was the Russian governments inability to balance its finances. A vicious circle of increased domestic and international borrowing at ever-higher interest rates ended when the supply of external financing dried up. Russia was forced to abandon its fixed exchange rate and reschedule its loan commitments. Sutela (2000) discusses the crisis and the events leading up to it. The collapse made it clear that an alternative regime was needed for monetary and exchange rate policies. The adopted (and current) regime might be described as a hybrid of inflation targeting and managed float. In its annual Guidelines for the Single State Monetary Policy, the CBR specifies the inflation target for the next year and includes a forecast range for growth of the ruble money stock (M2). Korhonen and Mehrotra (2007) list these inflation targets and money growth projections, as well as actual outcomes, for the period 19992006. The conduct of monetary policy in that period was not straightforward. The lack of monetary policy instruments and underdeveloped interbank markets compelled the CBR to rely heavily on foreign exchange interventions. The CBR announces annual ceilings for appreciation of the real ruble rate. During the postcrisis era, the fiscal authorities have helped Russian monetary policy in fighting inflation. Strong budget surpluses andespecially from 2004 onwardgrowth of the Stabilization Fund have helped to regulate liquidity in the economy. In an overview of nominal anchors in the Commonwealth of Independent States (CIS) countries, Keller and Richardson (2003) find that almost all CIS countries, even if they claim to have floating exchange rates, are heavily involved in managing the external values of their currencies. This reliance on exchange rate interventions as the main instrument of monetary policy is partly explained by high degrees of dollarization. The authors conclude that as inflation expectations decline and dedollarization gathers pace, other nominal anchors may become more optimal. Even so, direct inflation targeting appears to be out of reach for these central banks for a while. Previous research on money demand and prices in Russia includes a traditional money demand function approach along with estimations of various monetary policy rules. With a single exception, none of the studies reviewed here concentrate on the postcrisis period. Nikolic; (2000) estimates several models of inflation for Russia using data from the precrisis period. As one would expect, he finds that money Granger-causes inflation. However, the findings also suggest that inflation may Granger-cause growth of the broad money stock; to the extent that this reflects feedback effects in money demand, the central bank needs to account for such effects in its efforts to control the money stock. In addition, velocity was highly volatile before the crisis, compounding the difficulties in implementing stabilizing monetary policies. Oomes and Ohnsorge (2005) augment traditional money demand functions by including estimates of foreign cash holdings in the monetary aggregates. When testing for cointegration between the monetary aggregate, output variable, deposit interest rate, and exchange rate change, they find the most stable long-run relation for the specification in which both foreign currency deposits and cash holdings are included in the broad

Emerging Markets Finance & Trade

money aggregate. In the specifications with only the ruble money stock, the exchange rate change is always significant and a faster rate of depreciation decreases the demand for ruble balances. However, the model with the broad ruble money stock is reported to be the most unstable of all tested specifications. Vymyatnina (2006) studies the monetary transmission mechanism in Russia between July 1995 and September 2004. The hypothesis is that demand for real ruble balances depends on income (proxied by real total trade), opportunity cost of holding money (proxied by an interbank interest rate) and the exchange rate against the U.S. dollar. However, the estimated cointegration relation appears to be unstable, especially around the time of the 1998 crisis and from 2002 onward. Estimating a single model for the entire data sample fails to pick up the deepening monetization of the Russian economy of recent years (see Kim and Pirttil 2004). Though the relatively long sample obviously yields more observations, including the 1998 crisis complicates the interpretation of the obtained coefficients. Granville and Mallick (2006) estimate long-run relations between real money, the interest rate, and the exchange rate with monthly data from February 1992 to January 2005. According to their results, the exchange rate affects interest rates in Russia but does not directly affect inflation. However, the authors detect a break around the 1998 crisis, which prompts them to estimate the model for inflation and interest rate only for the precrisis period as well. The resulting coefficients are quite different from those obtained from the full sample, reinforcing our view that it is quite useful to concentrate on the postcrisis period. Esanov et al. (2006) estimate several monetary policy rules for Russia from the last quarter of 1993 to 2002. Although this is not exactly the same as estimating money demand functions, the exercise is relevant. The McCallum rule for monetary policy determines the growth of the monetary base by the target and actual rates of growth of the nominal GDP as well as the change in velocity. In addition, the authors use the U.S. dollar exchange rate in estimating both McCallum and Taylor monetary policy rules, as we do when estimating money demand functions for Russia. Esanov et al. (2006) also estimate a hybrid Ball rule for Russian monetary policy. They conclude that the McCallum rule, where monetary base is the policy instrument, best fits the data for the period. On one hand, the CBR has always paid attention to monetary aggregates, as evidenced by its public forecasts for M2 growth. This would make the result plausible. On the other hand, the exchange rate has figured prominently in CBR policy, both before and after the 1998 crisis. Furthermore, changes in velocity have meant that growth of the money stock has consistently overshot its target band in the postcrisis era. Yet the authors do not consider a policy rule under which the exchange rate is the policy variable. Interpreting the results is all the more challenged by the presence of large shifts in the levels of many variables following the 1998 financial crisis, and despite the authors best efforts to control for this with dummy variables. Vdovichenko and Voronina (2006) also look at different monetary policy rules in Russia, but they concentrate on the period between 2000 and 2003. Their analysis reveals that the CBR has continued to regard the stability of the exchange rate as very important. Also, the monetary base seems to be more relevant as a domestic monetary policy instrument than, say, the interest rate. The result concerning the exchange rate accords with our own results and we also leave out the interest rate in our estimations. In a CBR internal note, Ponomarenko (2007) estimates an error-correction money demand function for Russia using real M2 as the dependent variable between March 1999 and September 2006. In the long-run specification, the income variable is domestic

MarchApril 2010 9

absorption (household consumption, fixed capital formation, and government consumption). The income elasticity of money seems to be quite highover 2.5. However, the estimations are done in a single-equation framework, whereas we utilize a systems approach. Therefore, no impulse responses are reported either. From the studies reviewed here, it appears that a money demand relation can also be found for Russia, even with the complications from widespread dollarization and the August 1998 financial crisis. Nevertheless, it is apparent that these factors have introduced a large degree of uncertainty into the empirical results. Thus, we aim to provide more stable and policy-relevant results by restricting ourselves to data from the postcrisis period and a single monetary policy regime. Data The present study uses monthly data from January 1999 to December 2006, omitting the months immediately following the August 1998 crisis. This eight-year period, though short, provides enough observations for statistical inference about the postcrisis behavior of money demand in Russia. Admittedly, a cointegration analysis would benefit from a longer sample (see, e.g., Hakkio and Rush 1991). Nevertheless, the very nature of transition economies implies that a long consistent sample is not available.1 Moreover, economic theory focuses on long-run relations that are captured in the cointegration framework. In the main econometric model, we use monthly data on the nominal M2 ruble money stock, denoted mt; consumer price index, denoted pt; GDP indicator, denoted yt; and nominal ruble exchange rate against the U.S. dollar, denoted et.2 Figures depicting variables appear in Korhonen and Mehrotra (2007). We specify long-run real broad money demand in the Russian economy as

( m p)t = 1 yt + 2 t + 3et + ut ,

(1)

where all variables are in logarithms. Demand for real money balances is explained by real income, inflation (t, the month-to-month change in the consumer price index), and the exchange rate.3 We include the latter two variables to account for the opportunity cost of holding money. Interest rates are generally not considered appropriate indicators of monetary policy stance in Russia, given the low liquidity in the interbank markets. Pantyushin and Cherdantseva (2007) suggest that no interest rate affects inflation or investment dynamics. The use of a government bond (GKO) rate as a long-run interest rate is mitigated by the collapse of this market during the 1998 crisis and its subsequent lack of liquidity. Moreover, the CBR can control the exchange rate level to a considerable degree. As the Russian economy remains highly dollarized, the exchange rate potentially has a large effect on money demand. The consumer price index and the nominal exchange rate against the U.S. dollar are taken from the International Monetary Funds (IMF) International Financial Statistics (IFS) database. The nominal ruble stock comes from the CBR. The monthly GDP indicator comes from Russias statistical agency, Rosstat, and is based on output in five core sectors of the economy.4 Most previous studies on Russian money demand use industrial production to proxy income. Given the growing importance of services and construction in the Russian economy, we feel the GDP indicator better captures the dynamics of the overall economy. The order of integration of the series has important implications for the chosen estimation approach. We tested for unit roots with the NgPerron test. The maximum number for lags was set to 11, but the actual number of lags was determined by the Schwarz

10

Emerging Markets Finance & Trade

information criterion (lags were usually between 0 and 5). For real money, oil price, the exchange rate, inflation, and GDP, a unit root cannot be rejected for the series in levels, whether we include as deterministic variables a constant or a constant and trend.5 Therefore, treating the variables as integrated of order one, that is, nonstationary, would seem prudent, though in principle, stationary variables can also be accommodated in vector error-correction (VEC) systems. As our unit root tests suggest the series can be modeled as integrated of order one, it is possible that the series are driven by common stochastic trends. If so, they are said to be cointegrated. Thus, we prefer the VEC specification, as it allows us to distinguish between short-run and long-run relations. Theoretical relations, such as a money demand relation, are typically defined in terms of the levels of the series, whereas a lack of cointegration could lead to spurious regressions when the variables are integrated of an order higher than zero. We test for cointegration utilizing the test proposed by Saikkonen and Ltkepohl (2000), including a constant, linear trend, and seasonal dummies in the testing procedure, because some series trend over time. The oil price is not included, as it only appears in our estimated system as an exogenous variable. Table 1 reports the results. For our benchmark postcrisis sample, the SaikkonenLtkepohl test suggests the existence of one cointegration relation, as a cointegration rank of zero can be rejected at the 1 percent level, and a rank of one cannot be rejected at any conventional level. As a robustness test, we considered truncated samples 1999M22005M12 and 2000M12006M12, respectively. These alternative samples do not change our finding of a cointegrating rank of one, even at the 10 percent significance level. We continue with the assumption that a single cointegration relation exists among the system variables. Estimation Results and Their Interpretation As mentioned above, and like most recent research on money demand, we adopt a VEC model. Our cointegration tests justify our choice; they found that the variables share common stochastic trends that could plausibly be written in the form of a money demand relation (though this can only be tested in the context of system estimation). When we omit the deterministic terms and include exogenous variables, a reduced-form VEC model can be expressed as xt = xt 1 + 1 xt 1 + + p 1 xt p +1 + B0 zt + + Bp zt p + ut , (2)

where p denotes the order of the VEC model. K is the number of endogenous variables. xt = (x1t, ..., xKt) is a (K 1) random vector, and i are fixed (K K) coefficient matrices. zt are unmodeled stochastic variables and are thereby assumed to be exogenous. Bi are parameter matrices. The ut = (u1t, ..., uKt) is a K-dimensional white noise process with E(ut) = 0. When the variables are cointegrated, has reduced rank r = rk() < K. It can be expressed as = , where and are (K r) matrices that contain the loading coefficients and the cointegration vectors, respectively. We estimate the model with five lags in both the endogenous and exogenous variables. Our choice for the lag length is based on tests for misspecification and stability. As the observations for 1999M21999M7 are used as presample values, the effective estimation sample runs from 1999M8 to 2006M12. We include a constant, seasonal dummies, and a linear trend as the deterministic variables in the system. The trend is restricted in the cointegration relation. As the ruble-to-U.S. dollar exchange rate is a policy variable for

MarchApril 2010 11

Table 1. SaikkonenLtkepohl cointegration test

Sample
1999M22006M12

Lagged levels
1

Null hypothesis
r r r r r r r r r r r r = = = = = = = = = = = = 0 1 2 3 0 1 2 3 0 1 2 3

Test statistic
74.47*** 24.59 5.43 0.04 61.84*** 18.10 6.68 0.00 53.32*** 22.02 11.20 0.09

1999M22005M12

2000M12006M12

Notes: Deterministic terms include a constant, a linear trend, and seasonal dummies for all systems. *** significance at the 1 percent level.

the CBR, we only include it as unmodeled (and restricted) in the cointegration relation. It thus appears as one of the long-run determinants of money demand, without entering the short-run dynamics of the estimated system.6 Oil price in first differences is included as an exogenous variable. We estimate the cointegration relation by the simple two-step (S2S) procedure, which effectively utilizes feasible generalized least squares (FGLS). The S2S estimator displays substantially better small-sample properties than does the maximum likelihood (ML) estimator, as Brggemann and Ltkepohl (2005) show. The cointegration relation is estimated as (standard errors in parentheses):

( m p)t = 1.725 yt 4.593t 0.954et + 0.010t + ut . (0.324) (2.392) (0.135) (0.001)

(3)

The estimated coefficients have the signs that theory suggests. A higher level of output increases the demand for real money, whereas a higher opportunity cost of holding money leads to a fall in money demand. The coefficient on the inflation rate is weakly significant, just above the 5 percent level. Its relatively high absolute value could reflect the limited investment opportunities in the Russian economy outside real assets, as Banerji (2002) discusses. Our estimated income elasticity of money demand is higher than unity, reflecting increasing monetization of the Russian economy after the 1998 crisis. In their analysis of earlier surveys on money demand, Knell and Stix (2006) report that the average income elasticity of broad money for nonOrganization for Economic Cooperation and Development (OECD) countries is very close to unity, but the standard deviation of elasticity observations tends to be quite high, ranging from 0.36 to 0.53. Our estimate for income elasticity is not far off from the results found for other non-OECD countries. Finally, including a short-run money market interest rate that could plausibly assume the role of the own interest rate of broad money yields statistically insignificant estimates, for its long-run and adjustment coefficients.7 Thus, we omit it from the final model. Our estimated cointegration relation remains relatively robust to changes in the chosen VEC order, as Table 2 shows. Testing with a one-month ruble-to-U.S. dollar forward

12

Table 2. Estimated cointegration vectors with alternative lag lengths, spot rates, and forward contracts
Lags
4 5 6 7 6 2.079 1.725 1.989 1.589 1.268 0.371 0.324 0.337 0.232 0.276 5.555 4.593 5.680 6.692 17.275

Emerging Markets Finance & Trade

Method

Exchange rate

(y) SE(y)

()

SE()
2.599 2.392 2.863 2.060 3.215

(e)
1.029 0.954 1.016 1.014 1.090

SE(e)
0.153 0.135 0.145 0.098 0.089

S2S S2S S2S S2S S2S

Spot RUB/USD Spot RUB/USD Spot RUB/USD Spot RUB/USD One-month forward

Note: SE denotes standard errors.

MarchApril 2010 13

exchange rate offers additional insight, as the exchange rate sensitivity of the demand for Russian domestic broad money may vary using different exchange rates. Moreover, a forward rate measures exchange rate expectations, which should be important for the ruble money-holding sector, especially during the postcrisis period, when increased confidence in the domestic currency has triggered dedollarization. To our knowledge, previous studies have not used forward contracts to analyze Russian money demand.8 Table 2 reports the results. A caveat here is that the estimation sample starts only in 2000M7, due to data limitations (the forward data are available only from 2000M1 onward) and significant outliers in the beginning of 2000. We find that the forward exchange rate is statistically significant with the expected sign, and its coefficient estimate falls very close to that of the spot exchange rate in the benchmark estimation sample. However, our coefficient estimate for the inflation rate now increases to 17.275. We continue with the benchmark specification of using the spot ruble-to-U.S. dollar exchange rate (level), as the sample is longer, but note that the forward data bring about correctly signed and significant coefficient estimates for the exchange rate in the Russian money demand equation.9 Furthermore, through the covered interest rate parity, we are actually including the interest rate difference between Russia and the United States in the estimation, when we use the forward rate. The short-run parameters of the VEC system are estimated by FGLS. Model reduction was achieved by a procedure in which the parameter with the lowest t-value was checked and possibly eliminated. The final model only includes coefficients with a higher t-value than 1.67. This approach allows us to preserve degrees of freedom in our short sample without removing any statistically significant coefficients from the estimated system. Table 3 reports the loading coefficients, the entire short-run part of the system, and misspecification test results from the model. The adjustment coefficients on the cointegration relation ect1 in Table 3 are of interest, as they indicate the possible adjustment of our system toward long-run equilibrium. We find that the coefficient on the equation for real money (0.142) is statistically significant and negative, indicating that excess liquidity is corrected within our system. The speed of adjustment is relatively rapid. Excess liquidity in the Russian economy results in inflation pressure and higher real GDP, as indicated by the statistically significant adjustment coefficients in the inflation and output equations.10 For the short-run coefficients, there is feedback in the system between the money and price variables as they enter both equations. Similarly, we detect feedback between real money and output. Misspecification tests, also reported in Table 3, include the Lagrange multiplier (LM) test for autocorrelation, the JarqueBera test for nonnormality, and the multivariate vector autoregressive conditional heteroskedasticity (VARCH)-LM test for autoregressive conditional heteroskedasticity (ARCH) effects in model residuals. None of these tests suggest misspecification.11 The investigation of model stability is of special interest in money demand examinations. A stable money demand relation is a prerequisite for a meaningful specification of money growth targets, and could also be useful if concepts of excess liquidity, such as monetary overhang, are considered. We test for system stability by recursive eigenvalue tests, proposed by Hansen and Johansen (1999). The short-run parameters are concentrated out based on the full sample, and only the long-run part is estimated recursively. The standard errors to construct the 95 percent confidence intervals are estimated based on the full sample. The recursive eigenvalue remains relatively stable during the maximum period available for investigation. Additionally, the -statistic remains much smaller than the 5 percent critical value. The Chow forecast test, utilizing bootstrapped p-values obtained by 1,000 replications, also suggests model stability.

14

Emerging Markets Finance & Trade

Table 3. Loading coefficients, short-run part, and misspecification tests for vector error-correction model
Equation
ect1 D(mp)t1 D(mp)t2 D(mp)t3 D(mp)t4 D(mp)t5 Dyt1 Dyt2 Dyt3 yt4 yt5 t1 t2 t3 t4 t5 LM(5)=37.74 [0.77] JB1=0.35 [0.84] VARCHLM(5)=199.28 [0.15]

D (m p)
0.142 (0.025) 0.279 (0.079) 0.120 (0.074) 0.241 (0.079) 0.234 (0.049) 0.193 (0.043) 0.129 (0.046) 0.814 (0.290)

Dy
0.233 (0.025) 0.210 (0.078) 0.125 (0.037) 2.636 (0.517) 2.230 (0.583)

Dp
0.044 (0.013) 0.049 (0.017) 0.041 (0.011) 0.032 (0.007) 0.016 (0.008) 0.300 (0.066) 0.212 (0.067) 0.303 (0.078)

LM(4)=34.20 [0.55] LM(1)=8.82 [0.45] JB2=0.40 [0.82] JB3=0.08 [0.96]

Notes: Standard errors are in parentheses. The coefficients for the inflation rate are multiplied by 100. The LM-test for autocorrelation is conducted at one, four, and five lags. JB is the JarqueBera test for nonnormality for three individual system equations. VARCH-LM is the multivariate ARCH-LM test at five lags. The p-values for test statistics are in brackets. Constant, seasonal dummy, and exogenous variables are not displayed.

We test for a possible break date for every observation, commencing in 2003M2, which corresponds to the longest available sample in our case. For no observation can we reject parameter stability, even at the 10 percent level.12 We continue with the assumption that the estimated model displays stability over time. Our results of system stability for the ruble broad money stock contrast with Oomes and Ohnsorge (2005), and likely stem from the use of postcrisis data only in our estimation sample.13 To examine the forecasting performance of our money demand system in terms of the inflation rate, we estimated the model until 2006M5 and created out-of-sample forecasts for inflation seven months ahead. For the exogenous oil price variable, the actual outcomes during the period 2006M62006M12 were imposed. Figure 1 shows the forecast

MarchApril 2010 15

Figure 1. Out-of-sample inflation forecast

(bold dashed line), together with the actual outcome (bold solid line) and the confidence intervals (two thin dashed lines). All actual values fall inside the confidence intervals, supporting our estimated system. Shortening the estimation sample by one month and forecasting eight periods ahead, we find that all the actual outcomes similarly remain inside the confidence intervals. The forecast in the end of the period (2006M12) again falls quite close to the actual outcome. As our estimated money demand system displays stability over time, deviations of the money stock from the long-run equilibrium could, in principle, provide information about inflation pressures in the economy. Svensson (2000) points out that this real money gap, written as (mt pt) (mt* pt*), is equivalent to the negative of the price gap in P* models (see, e.g., Hallman et al. 1991; Tdter and Reimers 1994). Here, (mt* pt*) denotes the equilibrium money stock obtained by substituting the actual outcomes for real money, output, the inflation rate, and the exchange rate to the estimated long-run money demand equation. We examine the forecasting performance of the real money gap in terms of the inflation rate by testing for Granger causality, where causality is determined by a variables ability to improve the forecasts of another variable. We include the inflation rate (in first differences to ensure stationarity) together with the real money gap and seasonal dummies in the estimated system for the Granger causality test. This bivariate system, using one lag as suggested by the Schwarz information criterion, passes tests for autocorrelation and nonnormality, similarly to the estimated VEC model above.14 We find that the null hypothesis of Granger noncausality from the real money gap to prices can be weakly rejected; the test statistic is 3.505 (p = 0.06). Therefore, the real money gap estimated within our money demand system has some predictive power for future inflation. For our system with a cointegrating relation, we implement the standard impulse response analysis in the context of a structural VEC model, where contemporaneous restrictions are used to identify the system. The structural VEC model is expressed as
* Axt = * xt 1 + 1 xt 1 + + * 1 xt p +1 + B t . p

(4)

The structural shocks t are mutually uncorrelated. We assume that t (0, IK). In Equation (4), * and j* (j = 1, ..., p 1) are structural-form parameter matrices. A is an invertible matrix and allows for modeling of the instantaneous relations. The structural shocks are related to the reduced-form residuals by linear relations, specifically ut = A1 B t . (5)

16

Emerging Markets Finance & Trade

We identify the model by considering restrictions on the parameter matrix B, with our ordering of the variables specified as (m p), y, . In particular, we impose * * * B = 0 * 0 , 0 * *

(6)

where asterisks denote the unrestricted elements. Such an identification scheme effectively amounts to a Cholesky decomposition of the variancecovariance matrix, with the ordering of the variables specified as output, prices, and real money. Output is the most rigid variable in our system, as it does not react within the same month to shocks in prices or real money. In the model by Sims and Zha (2006), there are significant adjustment costs and inertia associated with real economic activity. Prices adjust slowly due to menu costs, or because they are changed at exogenously determined random intervals, as in Calvo (1983). Considering real money an indicator of monetary policy stance, the noncontemporaneous effect of real money on output and prices is consistent with most work on reduced-form modeling of monetary policy. This real money stock is allowed to react rapidly to any news about the economy, or indeed, any macroeconomic shocks that affect real output and prices in our system. Figure 2 illustrates impulse responses forty-eight months ahead. Hall 95 percent bootstrapped percentile confidence intervals with 1,000 replications were used. In our system, a permanent shock to real broad money leads to a statistically significant and permanent increase in the inflation rate. This can be justified by considering the permanent shock to real money as a permanently looser stance of monetary policy. This result is in line with the finding that the system variables are integrated of order one; some shocks should thus have permanent effects. The same real money shock leads to a permanent increase in the level of GDP, as expected. As theory predicts, an inflation shock leads to a decline in the real money stock, with a statistically significant effect. A shock to real GDP leads to an increase in the money stock, which is in line with higher transaction demand for broad money. Finally, the effects of permanent shocks stabilize in about 1.5 years in our system, supporting the idea that the sample is adequate for meaningful inference about long-run relations between the system variables. Conclusion and Policy Implications We have estimated different specifications for money demand in Russia. To our knowledge, such analysis in a VEC framework has not been applied solely to data from the period following the August 1998 financial crisis. While this approach limited the available data, it permitted us to concentrate on a period with a relatively homogenous monetary policy regime. The CBR has maintained fairly tight control of the nominal exchange rate, whereas inflation has decelerated. Nevertheless, the CBR has failed repeatedly to meet its inflation targets. Inflation seemed to plateau at around 10 percent in 2005, though quite recently (spring 2007), further declines have been registered. We find a stable money demand function for Russia between the years 1999 and 2006. Demand for real ruble balances dependsas theory predictson income (proxied by GDP) and the opportunity cost of holding money (proxied by inflation and the exchange rate). The estimated income elasticity of real money is somewhat higher than the average found for other non-OECD countries (Knell and Stix 2006). This result is

MarchApril 2010 17

Figure 2. Impulse responses

likely explained by remonetization of the Russian economy during our sample period. From a policy point of view, the finding of a stable money demand function for domestic currency is important, as it suggests that the domestic money stock can be used as an information variable about developments in the real economy. This result contrasts to previous literature covering the postcrisis era, such as Oomes and Ohnsorge (2005). We find that broad money shocks lead to higher inflation and that deviations from long-run equilibrium ruble money stock provide information about future inflation. Another insight of clear policy relevance is that exchange rate movements influence the demand for real money balances. Managing the ruble exchange rate can directly affect the remonetization of the Russian economy. These results are probably relevant for other emerging market countries where dollarization prevails. The existence of a stable money demand function at least partly vindicates the CBRs policy of linking money growth (in the form of money growth forecasts) and inflation targets. Of course, the practical implementation of such policy has been made more difficult by the registered decline in velocity. If the growth in ruble money stock is driven solely by remonetization, inspired by increased confidence in domestic currency, it is unlikely to lead to inflationary pressures. In our model, we capture this development with a linear trend, but we have the luxury of performing our analysis ex post. Given that our results also show that the rubles exchange rate remains an important determinant of money demand, the close attention paid by Russias monetary authorities to the ruble rate seems well justified. Notes
1. Cointegration techniques have been previously used for Russia for similar or shorter estimation samples (e.g., Choudhry 1998; Oomes and Ohnsorge 2005; Rautava 2004).

18

Emerging Markets Finance & Trade

2. When checking for the robustness of our results, we also use the one-month forward ruble/U.S. dollar exchange rate. 3. An alternative approach would be to replace the demand for real money balances with the ratio of domestic currency to foreign currency deposits. This would lead to examining currency substitution in the Russian economy, which is an important phenomenon in itself (for a recent study, see Harrison and Vymyatnina 2007 and the references therein). At the end of 2007, the ratio of ruble time deposits to foreign currency deposits was approximately 4:1. We focus instead on the link between the ruble money stock, inflation, and output, which are important to the conduct of domestic monetary policy. 4. The five core sectors are industrial production, retail trade, construction, transport, and agriculture. 5. For the first differences of our series, the null hypothesis of a unit root is always rejected at the 1 percent level, with the exception of inflation. The rapid disinflation may be behind this counterintuitive result. If we use the augmented DickeyFuller test, which has lower power in small samples, the null hypothesis of having a unit root is rejected at the 1 percent level. (The test statistic is 13.079 when a constant and a trend are included.) Detailed results from unit root tests are available from the authors upon request. 6. Our approach is identical to that of Kalra (1999), who includes the state-owned banks deposit interest rate as an unmodeled variable in the money demand cointegration relation for the transition economy of Albania. 7. Similarly, in their long-run demand relation for the ruble broad money stock, Oomes and Ohnsorge (2005) find that the coefficient on the deposit interest rate is insignificant. 8. Bofinger (2001) uses data on the exchange rates of four major currencies against the U.S. dollar to investigate the fit of a regression where future spot rates are explained by data on current forward contracts. He shows that when the exchange rate data are in differences, the fit of the model is quite poor compared to estimations where the data are in the form of nonstationary levels. Froot and Thaler (1990) provide an overview of the literature on the topic. For our purposes, the issue is whether forward data provide similar estimates in the money demand system as the spot rate with the perfect foresight assumption. 9. We also tested for the inclusion of the nominal effective exchange rate, but our estimates for the long-run money demand did not remain robust when this variable was used. The U.S. dollar remains the most widely used foreign currency in Russia in terms of economic transactions, savings accounts, and even in pricing. 10. The t-values of the adjustment coefficients in the inflation and output equations are 3.440 and 9.244, respectively. 11. In contrast, including either the money market, refinancing, or deposit interest rate leads to nonnormality in the estimated system. 12. The results from stability tests are available on request. 13. Using an estimation period from December 1996December 2006, thereby including the Russian crisis in the estimation sample, we cannot find a single model that passes both the tests for misspecification and stability. This holds for any tested lag length between 1 and 12. 14. In the LM test, there is very weak evidence of autocorrelation when four lags are used (p-value of 0.099).

References
Bahmani-Oskooee, M., and M.P. Barry. 2000. Stability of the Demand for Money in an Unstable Country: Russia. Journal of Post Keynesian Economics 22, no. 4: 619629. Banerji, A. 2002. Money Demand. In Country Report no. 02/75, International Monetary Fund, Washington, DC. Bofinger, P. 2001. Monetary Policy: Goals, Institutions, Strategies, and Instruments. New York: Oxford University Press. Brggemann, R., and H. Ltkepohl. 2005. Practical Problems with Reduced-Rank ML Estimators for Cointegration Parameters and a Simple Alternative. Oxford Bulletin of Economics and Statistics 67, no. 5: 673690. Calvo, G. 1983. Staggered Prices in a Utility-Maximizing Framework. Journal of Monetary Economics 12, no. 3: 383398.

MarchApril 2010 19 Choudhry, T. 1998. Another Visit to the Cagan Model of Money Demand: The Latest Russian Experience. Journal of International Money and Finance 17, no. 2: 355376. Esanov, A.; C. Merkel; and L. Vinhas de Souza. 2006. Monetary Policy Rules for Russia. In The Periphery of the Euro: Monetary and Exchange Rate Policy in CIS Countries, ed. L. Vinhas de Souza and P. De Lombaerde, pp. 145168. Hant, UK: Ashgate. Froot, K.A., and R.H. Thaler. 1990. Anomalies: Foreign Exchange. Journal of Economic Perspectives 4, no. 3: 179192. Granville, B., and S. Mallick. 2006. Does Inflation or Currency Depreciation Drive Monetary Policy in Russia? Research in International Business and Finance 20, no. 2: 163179. Hakkio, C.S., and M. Rush. 1991. Cointegration: How Short Is the Long Run? Journal of International Money and Finance 10, no. 4: 571581. Hallman, J.J.; R.D. Porter; and D.H. Small. 1991. Is the Price Level Tied to the M2 Monetary Aggregate in the Long Run? American Economic Review 81, no. 4: 841858. Hansen, H., and S. Johansen. 1999. Some Tests for Parameter Constancy in Cointegrated VAR Models. Econometrics Journal 2, no. 2: 306333. Harrison, B., and Y. Vymyatnina. 2007. Currency Substitution in a Dedollarizing Economy: The Case of Russia. Discussion Paper 3/2007, Bank of Finland, Institute for Economies in Transition, Helsinki. Kalra, S. 1999. Inflation and Money Demand in Albania. Russian and East European Finance and Trade 35, no. 6: 82105. Keller, P.M., and T.J. Richardson. 2003. Nominal Anchors in the CIS. Working Paper no. 03/179, International Monetary Fund, Washington, DC. Kim, B.-Y., and J. Pirttil. 2004. Money, Barter, and Inflation in Russia. Journal of Comparative Economics 32, no. 2: 297314. Knell, M., and H. Stix. 2006. Three Decades of Money Demand Studies: Differences and Similarities. Applied Economics 38, no. 7: 805818. Korhonen, I., and A. Mehrotra. 2007. Money Demand in Postcrisis Russia: Dedollarization and Remonetization. Discussion Paper 14/2007, Bank of Finland, Institute for Economies in Transition, Helsinki. Nikolic;, M. 2000. Money GrowthInflation Relationship in Post-Communist Russia. Journal of Comparative Economics 28, no. 1: 108133. Oomes, N., and F. Ohnsorge. 2005. Money Demand and Inflation in Dollarized Economies: The Case of Russia. Journal of Comparative Economics 33, no. 3: 462483. Pantyushin, V., and O. Cherdantseva. 2007. Choosing Right . . . and the Chicken-Egg Dilemma. Renaissance Capital Research Report, July 19, 2007, Moscow. Ponomarenko, A. 2007. Modeling Money Demand in Russia. Central Bank of Russia, Moscow. Rautava, J. 2004. The Role of Oil Prices and the Real Exchange Rate in Russias Economy: A Cointegration Approach. Journal of Comparative Economics 32, no. 2: 315327. Saikkonen, P., and H. Ltkepohl. 2000. Testing for the Cointegrating Rank of a VAR Process with Structural Shifts. Journal of Business and Economic Statistics 18, no. 4: 451464. Sims, C.A., and T. Zha. 2006. Does Monetary Policy Generate Recessions? Macroeconomic Dynamics 10, no. 2: 231272. Sutela, P. 2000. The Financial Crisis in Russia. In Global Financial Crisis: Lessons from Recent Events, ed. J. Bisignano, W. Hunter, and G. Kaufman, pp. 6376. Norwell, MA: Kluwer. Svensson, L.E.O. 2000. Does the P* Model Provide Any Rationale for Monetary Targeting? German Economic Review 1, no. 1: 6981. Tdter, K.-H., and H.-E. Reimers. 1994. P-Star as a Link Between Money and Prices in Germany. Weltwirtschaftliches Archiv 130, no. 2: 273289. Vdovichenko, A.G., and V.G. Voronina. 2006. Monetary Policy Rules and Their Application in Russia. Research in International Business and Finance 20, no. 1: 145162. Vymyatnina, Y. 2006. Monetary Policy Transmission and CBR Monetary Policy. In Return to Growth in CIS Countries: Monetary Policy and Macroeconomic Framework, ed. L. Vinhas de Souza and O. Havrylyshyn, pp. 2339. Berlin: Springer.

To order reprints, call 1-800-352-2210; outside the United States, call 717-632-3535.

Copyright of Emerging Markets Finance & Trade is the property of M.E. Sharpe Inc. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.