I. Introduction

The relationship between money growth and inflation in Russia has recently seemed difficult to understand. In 2003, ruble broad money growth accelerated from 35 to 50 percent, while headline inflation fell markedly, from 15 to 12 percent. Since 2004, the opposite phenomenon has occurred: ruble broad money growth slowed substantially, from almost 60 percent to around 25 percent by mid-2005, while inflation has picked up (Figure 1). A substantial part of the movements in headline inflation were due to temporary or seasonal factors, as indicated by the fact that core inflation was roughly stable.² Nevertheless, the stability of core inflation is difficult to reconcile with these sharp movements in money growth.

View Full Size

Core and Headline CPI and Ruble Broad Money

(12-month percent change)

Citation: IMF Working Papers 2005, 144; 10.5089/9781451861631.001.A001

• Download Figure
• Download figure as PowerPoint slide

In this paper, we argue that taking into account foreign cash holdings is crucial for understanding the relationship between money growth and inflation in Russia. Since the early 1990s, foreign currency, mainly the U.S. dollar, has generally served all standard money functions in Russia.³ It is important, therefore, to include foreign cash holdings outside the banking system in the definition of money. While data on foreign cash holdings are unavailable for most countries, such data are available for Russia, and are described in Appendix I. Although the exact amount of foreign cash holdings is subject to considerable uncertainty, it is clear that, following a period of rapid dollarization (i.e., an increase in the share of foreign cash and foreign currency deposits) until end-1999, Russia has experienced the opposite phenomenon—that is, dedollarization—in recent years.⁴ In particular, Figure 2 suggests that the share of foreign cash holdings in overall money demand has declined substantially, and has largely been offset by an increase in the share of ruble deposits and, to a lesser extent, ruble cash.

View Full Size

Effective Broad Money Components

(In percent)

Citation: IMF Working Papers 2005, 144; 10.5089/9781451861631.001.A001

• Download Figure
• Download figure as PowerPoint slide

The dynamics of dollarization appear to be related closely to movements in the exchange rate. As Figure 3 shows, the growth in ruble broad money (defined as ruble currency plus deposits, or M2) slowed during periods of ruble depreciation (e.g., during 2001 and 2002), and accelerated during periods of ruble appreciation (e.g., during 2003). A likely explanation for this strong correlation is that ruble depreciation increases the returns on holding foreign currency and, therefore, gives incentives to dollarize, while ruble appreciation increases the returns on holding ruble currency and, therefore, gives incentives to dedollarize. To the extent that dollarization and dedollarization represent mere portfolio shifts, it is perhaps not surprising that the growth in effective broad money (defined as M2 plus foreign currency deposits plus foreign cash holdings) has been much more stable than ruble broad money growth.

View Full Size

Monetary Aggregates (12-month percent change), and Exchange Rate (Ruble per U.S. Dollar)

Citation: IMF Working Papers 2005, 144; 10.5089/9781451861631.001.A001

• Download Figure
• Download figure as PowerPoint slide

Previous studies of money demand in Russia, which did not include an estimate of foreign cash holdings, have found it difficult to find a stable money demand function. Banerji (2002) estimates a demand function for ruble broad money in Russia for the period June 1995– March 2001 and finds that VAR tests for the presence of a single cointegrating vector, which is one measure of money demand stability, “did not yield sensible or robust results.”⁵ Choudhry (1998) estimates money demand in Russia for the earlier hyperinflation period January 1992–September 1994 and also fails to find evidence for a stationary long-run relationship between real ruble money balances (ruble broad money and ruble currency) and inflation.⁶ Bahmani-Oskooee and Barry (2000) obtain a stationary relationship for the period January 1991–June 1997, but, nevertheless, find evidence for instability of real ruble broad money demand in Russia. Studying a similar period, January 1992–July 1998, Nikolić (2000) finds that the relationship between broad money and inflation is “unstable and sensitive to changes taking place in the new economic and institutional environment” (p. 131).

In this paper, we show that including foreign cash holdings in the definition of money improves the stability of the money demand function. For the period April 1996–January 2004, we find that monetary aggregates that exclude foreign cash holdings are significantly negatively dependent on the nominal bilateral U.S. dollar depreciation rate, confirming that the U.S. dollar has been an important substitute for ruble broad money. It is perhaps not surprising, therefore, that the demand for effective broad money, which includes foreign cash holdings, does not significantly depend on depreciation, because it is unaffected by currency substitution. Additional indications that the effective broad money demand function is the most stable are that it exhibits the smallest standard errors, constant parameter estimates, and the strongest evidence for uniqueness of its cointegrating vector.

The construction of a broader, more stable, monetary aggregate can help us to better understand, and possibly predict, inflation. Using a standard approach for estimating a “long-run” inflation equation, we find that, over our sample period, nominal effective depreciation accounts for roughly 50 percent, the growth in unit labor costs for 40 percent, and utility price growth for 10 percent of inflation. Deviations from this long-run relationship can be quite persistent—they can take up to 12 months to be corrected. In addition, the short-run dynamics of inflation seem to depend importantly on effective broad money. One indication of this is that an excess supply of effective broad money does appear to be inflationary, while the excess supplies of narrower monetary aggregates do not. A second indication is that changes in effective broad money growth have the strongest and most persistent effect on short-run inflation.

Our results should be interpreted with some caution, since they depend crucially on the accuracy of the data and the appropriateness of the econometric approach. First, our estimate of foreign currency in circulation is subject to considerable uncertainty (Appendix I). Second, the strength of the short-term impact of effective broad money on inflation may be overestimated because of valuation effects (Appendix II). Third, our “long-run” cointegration results should be treated cautiously because of the relatively short time series and our ensuing choice to use monthly data and dummies for the August 1998 crisis, which may not have been sufficient to entirely remove the structural break in our series. Finally, while our methodology follows that of several other studies in the literature, the robustness of our approach needs to be tested in further research using other data sources, additional variables, and different econometric specifications.

The results of this study may be relevant for other dollarized transition economies. As Table 1 shows, a number of transition economies, mostly member countries of the Commonwealth of Independent States (CIS), experienced rapid money growth in 2003 while inflation remained relatively low. In addition, several recent studies on inflation in transition economies have found a weak link between inflation and the growth of monetary aggregates that exclude foreign cash holdings.⁷ It may, therefore, be useful to study whether the inclusion of an estimate of foreign cash holdings can help improve the stability of money demand functions for these transition economies as well.

Table 1.

Selected Transition Economies: Money Growth and Inflation, 2003

(Average annual percent growth)

Source: IMF, International Financial Statistic database.

¹Defined as currency outside banks, demand deposits, and time and savings deposits.

²Defined as M2 plus foreign currency deposits at deposit money banks.

Table 1.

Selected Transition Economies: Money Growth and Inflation, 2003

(Average annual percent growth)

	Inflation	M21	Broad Money2
Azerbaijan	2.2	28.0	30.0
Belarus	28.4	71.1	56.8
Bulgaria	2.3	20.3	16.8
Kazakhstan	6.4	55.1	29.2
Kyrgyz Republic	2.7	33.5	33.4
Russia	13.7	51.6	39.7
Ukraine	5.2	43.5	47.2

Source: IMF, International Financial Statistic database.

¹Defined as currency outside banks, demand deposits, and time and savings deposits.

²Defined as M2 plus foreign currency deposits at deposit money banks.

View Table

Table 1.

Selected Transition Economies: Money Growth and Inflation, 2003

(Average annual percent growth)

	Inflation	M21	Broad Money2
Azerbaijan	2.2	28.0	30.0
Belarus	28.4	71.1	56.8
Bulgaria	2.3	20.3	16.8
Kazakhstan	6.4	55.1	29.2
Kyrgyz Republic	2.7	33.5	33.4
Russia	13.7	51.6	39.7
Ukraine	5.2	43.5	47.2

Source: IMF, International Financial Statistic database.

¹Defined as currency outside banks, demand deposits, and time and savings deposits.

²Defined as M2 plus foreign currency deposits at deposit money banks.

II. Russia’s Inflation Experience During the Past Decade

Following a period of rapid ruble depreciation and hyperinflation in the early 1990s, tight monetary and fiscal policies implemented in 1995 helped to bring down 12-month inflation from over 200 percent at end-1994 to about 20 percent at end-1996. Disinflation efforts were supported by a favorable external environment. However, short-term capital inflows and official debt relief exerted upward pressure on the exchange rate. By attempting to stem ruble appreciation with unsterilized foreign exchange interventions, the CBR slowed the disinflation process somewhat. To discourage speculation for ruble appreciation, an exchange rate band was introduced from July 1995, which helped stabilize exchange rate and inflation expectations further (Figure 4).

View Full Size

Headline CPI and Ruble-Dollar Exchange Rate

(12-month percent change)

Citation: IMF Working Papers 2005, 144; 10.5089/9781451861631.001.A001

• Download Figure
• Download figure as PowerPoint slide

Inflation increased dramatically again following the August 1998 financial crisis. The seeds for the crisis had been sown by weak fiscal and external fundamentals, combined with contagion effects from the Asian crisis.⁸ In August 1998, the authorities announced a unilateral moratorium on debt payments and widened the trading band of the ruble, implying a de facto devaluation. In September 1998, the exchange rate depreciated by more than 100 percent, the immediate effect of which was a surge in monthly inflation from 3.7 percent in August to 38 percent in September 1998, and an increase in 12-month inflation from 5.6 percent in July 1998 to 126 percent by July 1999. The effect of depreciation to inflation was particularly strong given that many prices had been indexed in dollars.

Since 1999, the authorities have pursued a relatively unambitious disinflation path. A favorable external environment, in particular a recovery of oil and metals prices and ensuing current account surpluses, as well as resumed capital inflows from 2001, allowed the CBR to accumulate reserves. Since 2000, the CBR has been fighting pressures for ruble appreciation through largely unsterilized foreign exchange purchases. As a result, monetary aggregates have been growing rapidly, and further disinflation is proving difficult.

Effective broad money has grown broadly in line with inflation and exchange rate developments. As Figure 5 shows, effective broad money growth accelerated sharply during (and partly anticipating) the 1998 crisis,⁹ and decelerated again as inflation came down. While this is mostly a reflection of the revaluation of dollar-denominated assets during ruble depreciations and appreciations, more detailed calculations in Appendix II show that it also reflects a volume effect. Moreover, as we argue in Appendix II, the fact that changes in the exchange rate affect effective broad money may make it a more useful measure for explaining, and possibly predicting, inflation.

View Full Size

Headine CPI and Monetary Aggregates

(12-month percent change)

Citation: IMF Working Papers 2005, 144; 10.5089/9781451861631.001.A001

• Download Figure
• Download figure as PowerPoint slide

III. Analytical Framework

We model the determinants of inflation in Russia by combining two main theories: the markup theory of inflation and the monetary theory of inflation. The markup theory of inflation goes back to Duesenberry (1950), while the monetary theory of inflation is typically associated with Friedman and Schwartz (1963). The markup theory has often been used as a framework for estimating the long-run determinants of inflation, while the monetary theory is mostly used to model short-run inflation dynamics.

According to the markup theory of inflation, the long-run domestic price level is a markup over total unit costs. Following De Brouwer and Ericsson (1998), we can write:

where P is the consumer price index, µ is the markup, ULC is the unit labor cost (average wage cost per unit of output), P_im measures the price of imported inputs, and P_ut is an index of utility prices, including energy and electricity. The elasticities of the CPI with respect to each input price are denoted by α, β, and γ, respectively, and are hypothesized to be positive and constant. Moreover, it is typically assumed that P is linearly homogenous (that is, raising the cost of each input by x percent should lead to an increase in consumer prices by x percent), which generates the testable hypothesis α+β+γ=1. The term µ−1, then, equals the percentage markup of consumer prices over production costs. Similar markup models of inflation have been used, among others, by Lissovolik (2003) for Ukraine, by Kuijs (2002) for the Slovak Republic, and by Sekine (2001) for Japan.

According to the monetary theory of inflation, inflation is driven by excess money supply. Several studies have suggested that, in the years following price liberalization in 1992, inflation in Russia had mainly monetary roots, in that ruble money growth appeared to be strongly correlated with inflation.¹⁰ This suggests that, during the hyperinflation period of the early 1990s, money demand growth was negligible compared with money supply growth. In principle, however, inflationary pressures are naturally expected to arise only when money supply exceeds money demand.

We employ a two-stage estimation method that allows us to take into account both theories of inflation. In the first stage, we estimate a long-run markup equation for inflation without taking into account excess money supply (since excess money should equal zero in the long run) and we separately estimate a long-run equation for money demand, using the Johansen cointegration approach in both cases. In the second stage, we then combine the two long-run relationships in one equation that takes into account deviations from the long-run equations (i.e., the equilibrium correction terms), so as to determine the short-run dynamics of inflation.¹¹

A. Long-Run Inflation

To estimate a long-run inflation equation for Russia, we calculate unit labor costs using industrial wage, employment, and output data, we measure import price growth by nominal effective exchange rate depreciation, and we estimate the growth in utility costs by the growth of the “paid services” component of the CPI.¹² Unit labor cost are calculated as the industrial wage bill (i.e., industrial wages times industrial employment) divided by real industrial output.¹³ The nominal effective exchange rate is a trade-weighted average of exchange rates, and, therefore, is a better proxy for import prices than bilateral exchange rates.¹⁴ The “paid services” component of the CPI includes public transportation, housing, telephone subscription, electricity, water, sewage, and gas—i.e., most of the items that are also excluded from core inflation.¹⁵ An alternative, and possibly preferable, approach would be to use core inflation as the dependent variable, but this option was not available to us since Russian data on core inflation are only available from January 1999.

We can write the long-run equation for inflation as a loglinear function of unit labor costs, depreciation, and utility prices:

$\begin{array}{ccc}P=\ln (\mu )+\alpha \cdot \mathit{\text{ulc}}+\beta \cdot \mathit{\text{neer}}+\gamma \cdot p_{\mathit{\text{ut}}},& & (2)\end{array}$

where p denotes the logarithm of the CPI index, neer denotes the logarithm of the nominal effective exchange rate index, ulc the logarithm of the unit labor cost index, and p_ut the logarithm of the price index for paid services (utilities). In order to test for cointegration between these variables, it is necessary for all variables to be I(1), i.e., nonstationary in levels but stationary in differences. However, standard unit root tests, reported in Appendix III, cannot reject (at the 1 percent level) the hypothesis that all variables are nonstationary in differences, not even when we use 12-month differences to reduce seasonal effects.¹⁶ However, monthly changes in the year-on-year growth rates are found to be stationary. Although unit root tests have weak power in short time series, we consider these results sufficiently reliable to estimate the following cointegration vector:

$\begin{array}{ccc}{\Delta }p=\alpha \cdot {\Delta }\mathit{\text{ulc}}+\beta \cdot {\Delta }\mathit{\text{neer}}+\gamma \cdot {\Delta }{p}_{\mathit{\text{ut}}},& & (3)\end{array}$

where the ∆ signs refer to 12-month differences. That is, we assume that inflation depends for α percent on unit labor cost growth, for β percent on the nominal effective depreciation rate, and for γ percent on utility price inflation.

Johansen cointegration tests generally confirm the hypothesis of a unique cointegrating vector for the period April 1996–January 2004. As Table A2 in Appendix III shows, the null hypothesis of no cointegration (r = 0) is strongly rejected, while the null hypothesis that there is at most one cointegration test (r ≤ 1) cannot be rejected by three out of four cointegration tests (and for the fourth test only at the 5 percent level). Note that these tests include a small sample adjustment, which is important given the weak power of cointegration test in short time series.

The estimated cointegration relationship, normalized for inflation, suggests that long-run inflation depends for roughly 50 percent on nominal effective depreciation, for roughly 40 percent on unit labor cost growth, and for roughly 10 percent on increases in utility prices (Table 2). The estimate of the pass-through effect from nominal effective depreciation is relatively precise and robust, and close to the 0.5–0.6 pass-through estimate found by Stavrev (2003) for the nominal ruble-U.S. dollar depreciation rate.¹⁷ Without imposing the restriction that the coefficients should add up to unity, the coefficient on utility price growth is larger, and the coefficient on ULC is smaller. However, when we reestimate the equation imposing the linear homogeneity hypothesis (α+β+γ=1), a formal likelihood-ratio test shows that this hypothesis cannot be rejected at the 5 percent level. We, therefore, take the restricted long-run inflation equation to be the more economically meaningful one, although one should keep in mind that the coefficient on utility price growth is a residual.

Table 2.

Estimated Long-Run Cointegration Equation for Inflation, April 1996–January 2004¹

Note: ** denotes significance at the 1 percent level.

¹The regression includes three lags, a constant, a trend, and four dummies: for the disinflation period in early 1996, the 1998 crisis, the August 1999 recovery from the exchange rate overshooting in August 98, and October 2000 (for 4/96 to 7/96, 8/98 to 9/98, 8/99, 10/00).

²The coefficient on utility price growth is a residual.

Table 2.

Estimated Long-Run Cointegration Equation for Inflation, April 1996–January 2004¹

		Unit Labor Cost Growth	Nominal Effective Depreciation	Utility Price Growth	Sum of Coefficients
Unrestricted	Coefficient Standard error	0.253 ** 0.097	0.477 ** 0.019	0.457 ** 0.140	1.187
Restricted	Coefficient Standard error	0.397 ** 0.103	0.494 ** 0.020	0.109 2 …	1.000
LR test of restriction: χ2(1) =3.0548 [0.0805]

Note: ** denotes significance at the 1 percent level.

¹The regression includes three lags, a constant, a trend, and four dummies: for the disinflation period in early 1996, the 1998 crisis, the August 1999 recovery from the exchange rate overshooting in August 98, and October 2000 (for 4/96 to 7/96, 8/98 to 9/98, 8/99, 10/00).

²The coefficient on utility price growth is a residual.

View Table

Table 2.

Estimated Long-Run Cointegration Equation for Inflation, April 1996–January 2004¹

		Unit Labor Cost Growth	Nominal Effective Depreciation	Utility Price Growth	Sum of Coefficients
Unrestricted	Coefficient Standard error	0.253 ** 0.097	0.477 ** 0.019	0.457 ** 0.140	1.187
Restricted	Coefficient Standard error	0.397 ** 0.103	0.494 ** 0.020	0.109 2 …	1.000
LR test of restriction: χ2(1) =3.0548 [0.0805]

Note: ** denotes significance at the 1 percent level.

¹The regression includes three lags, a constant, a trend, and four dummies: for the disinflation period in early 1996, the 1998 crisis, the August 1999 recovery from the exchange rate overshooting in August 98, and October 2000 (for 4/96 to 7/96, 8/98 to 9/98, 8/99, 10/00).

²The coefficient on utility price growth is a residual.

The estimated coefficients of the (restricted) long-run inflation equation are robust and stable over time. All coefficients, and in particular the coefficient on nominal effective depreciation, appear broadly robust to the choice of lag length (using two instead of three lags), the measure of unit labor cost (using industrial wage growth gives similar results) and the measure of import price growth (using the bilateral depreciation rate against the U.S. dollar gives similar results as using nominal effective depreciation). Moreover, the coefficient estimates remain stable even when the equation is recursively estimated, i.e., when we sequentially add new monthly observations, starting from January 2000.¹⁸

B. Long-Run Money Demand

We estimate long-run money demand using a standard specification that is consistent with a range of monetary theories:

$\begin{array}{ccc}\frac{M^d}{p}=f(Y,\overline{R}),& & (4)\end{array}$

where M^d is the demand for a particular monetary aggregate, P is the consumer price index, Y is a scale variable measuring the real level of economic activity, and $\overline{R}$ is a vector representing the rates of return on alternative assets (i.e., the opportunity costs of holding money). To measure Y, we use monthly data on industrial production as a proxy for monthly GDP.¹⁹

To measure the rate of return on alternative assets, we use two different opportunity costs. The first opportunity cost is the nominal ruble deposit rate, which is a weighted average of interest rates on deposits with different maturities. While it is more common to use the T-bill rate, we could not use this for our sample period because of lack of data.²⁰ We apply this interest rate to monetary aggregates that also include deposits because we expect the cash component of monetary aggregates to dominate in Russian data. The second opportunity cost is the (expected) nominal ruble-dollar depreciation rate, that is, the rate of return on holding U.S. dollars.²¹ While the inflation rate could also be considered as an opportunity cost, i.e., the return on holding goods, the high correlation between inflation and depreciation in Russia prevented us from including both in the money demand equation at the same time.²² Hence, we chose to include only the depreciation rate because the depreciation rate may have been a more important opportunity cost than the inflation rate because (1) the depreciation rate is easier to monitor by the population than the inflation rate, and (2) anecdotal evidence suggests that U.S. dollars have been a more important alternative asset than durable goods.²³

While money supply can temporarily exceed money demand, or vice versa, it is assumed that prices in Russia are sufficiently flexible so as to ensure money market equilibrium in the long run. Setting M^d = M^s = M, and using lower cases for logarithms, we can thus write the long-run money demand function in log-linear form:

$\begin{array}{ccc}m-p={\beta }_0+{\beta }_{1}y+{\beta }_{2}i+{\beta }_{3}e,& & (5)\end{array}$

where i is the 12-month nominal ruble deposit rate, and e is the 12-month nominal ruble-U.S. dollar depreciation rate. Since each variable is nonstationary in levels, and stationary in (one-month) differences (Table A1, Appendix III), we can estimate a cointegrating relationship between the four variables.²⁴ The residual of this long-run relationship shows the difference between money supply and money demand, i.e., the monetary policy stance.

In order to determine which monetary aggregate provides the most stable money demand function, we estimate M^d by five different aggregates: ruble currency in circulation, ruble narrow money, ruble broad money, broad money, and effective broad money. Following Feige (2002), we define effective broad money (EBM) as the sum of ruble currency in circulation outside the banking system (RCC), ruble demand deposits (RDD), ruble time and savings deposits (RTD), foreign currency deposits (FCD) and, most importantly, an estimate of foreign currency in circulation (FCC). Our five monetary aggregates are defined as follows:

Ruble Currency in Circulation	= RCC
Ruble Narrow Money (RNM)	= RCC+RDD
Ruble Broad Money (RBM)	= RCC+RDD+RTD
Broad Money (BM)	= RCC+RDD+RTD+FCD
Effective Broad Money (EBM)	= RCC+RDD+RTD+FCC

View Table

Ruble Currency in Circulation	= RCC
Ruble Narrow Money (RNM)	= RCC+RDD
Ruble Broad Money (RBM)	= RCC+RDD+RTD
Broad Money (BM)	= RCC+RDD+RTD+FCD
Effective Broad Money (EBM)	= RCC+RDD+RTD+FCC

Cointegration tests suggest that the money demand function for effective broad money is the most stable. There is evidence for a unique cointegrating vector for effective broad money, even when this is calculated at fixed accounting exchange rates, so as to exclude the valuation effect described in Appendix II. However, the hypothesis of no cointegration can hardly be rejected for the two narrowest monetary aggregates, as Table A2 in Appendix III shows. However, since there is at least some evidence for a unique cointegration vector for all monetary aggregates (although for ruble narrow money only for one of the cointegration tests, and only at the 5 percent level), we do estimate long-run cointegration equations for all five definitions of money.

The estimated coefficients of the long-run money demand equations have the anticipated signs and are similar in most respects. As Table 3 shows, the coefficient estimate for industrial output, which measures the income elasiticty or transactions demand for money, is significant in all money equations. While the estimated income elasticity is in most cases not significantly different from unity, consistent with the quantity theory of money, it does exceed unity in all equations, which could possibly be explained by monetization, e.g., a decrease in barter transactions as the country grew richer..The coefficient estimate on the deposit rate, which serves as a general opportunity cost for holding money, is between −0.006 and −0.01, and is highly significant in all cases except for ruble broad money.²⁵ Since the deposit rate is measured in percentage points, this means a semi-elasticity of −0.6 percent to −1, meaning that an increase in the deposit rate by 1 percentage point leads to a decrease in money demand by 0.6 to 1 percent. The significance of the coefficients is broadly robust to the use of two lags (instead of three) and the inclusion of other variables (e.g., barter and the three-month U.S. treasury bill as an additional measure of opportunity cost).

Table 3.

Estimated Long-Run Money Demand Equations, April 1996–January 2004¹

Notes: ** denotes significance at the 1 percent level, and * denotes significance at the 5 percent level.

¹The regression includes three lags, a constant, seasonal dummies, a dummy for August-September 1998, and a dummy for August-September 1999.

Table 3.

Estimated Long-Run Money Demand Equations, April 1996–January 2004¹

			Industrial Production	Deposit Rate	Depreciation
Ruble currency in circulation
	Unrestricted	Coefficient Standard error	2.087 ** 0.587	-0.009 ** 0.004	-0.005 ** 0.002
Ruble narrow money
	Unrestricted	Coefficient Standard error	1.319 ** 0.414	-0.010 ** 0.003	-0.004 ** 0.001
Ruble broad money
	Unrestricted	Coefficient Standard error	3.159 ** 0.689	-0.004 0.005	-0.004 ** 0.002
Broad money
	Unrestricted	Coefficient Standard error	1.853 ** 0.326	-0.006 ** 0.002	-0.003 ** 0.001
Effective broad money
	Unrestricted	Coefficient Standard error	1.216 ** 0.126	-0.008 ** 0.001	0.001 0.000

Notes: ** denotes significance at the 1 percent level, and * denotes significance at the 5 percent level.

¹The regression includes three lags, a constant, seasonal dummies, a dummy for August-September 1998, and a dummy for August-September 1999.

View Table

Table 3.

Estimated Long-Run Money Demand Equations, April 1996–January 2004¹

			Industrial Production	Deposit Rate	Depreciation
Ruble currency in circulation
	Unrestricted	Coefficient Standard error	2.087 ** 0.587	-0.009 ** 0.004	-0.005 ** 0.002
Ruble narrow money
	Unrestricted	Coefficient Standard error	1.319 ** 0.414	-0.010 ** 0.003	-0.004 ** 0.001
Ruble broad money
	Unrestricted	Coefficient Standard error	3.159 ** 0.689	-0.004 0.005	-0.004 ** 0.002
Broad money
	Unrestricted	Coefficient Standard error	1.853 ** 0.326	-0.006 ** 0.002	-0.003 ** 0.001
Effective broad money
	Unrestricted	Coefficient Standard error	1.216 ** 0.126	-0.008 ** 0.001	0.001 0.000

Notes: ** denotes significance at the 1 percent level, and * denotes significance at the 5 percent level.

¹The regression includes three lags, a constant, seasonal dummies, a dummy for August-September 1998, and a dummy for August-September 1999.

All measures of money demand that exclude foreign currency in circulation are strongly negatively dependent on the nominal depreciation rate, suggesting that foreign cash has been an important substitute for domestic money and possibly even for foreign currency deposits.²⁶ For almost all measures of money demand, the semi-elasticity with respect to depreciation is estimated at approximately 0.4 percent, which implies that an increase in the depreciation rate by 1 percentage point leads to a decrease in money demand by 0.4 percent. For ruble broad money, for example, this implies that about one sixth of the growth in real ruble broad money in 2003 was accounted for by the ruble appreciation.²⁷ While broad money appears to be somewhat less sensitive to the depreciation rate with a semi-elasticity of 0.3 percent, this semi-elasticity is not significantly different from 0.4. As predicted, effective broad money demand, which includes foreign cash holdings, does not significantly depend on the depreciation rate.

Effective broad money is found to constitute the most stable money demand function, while the equation for ruble broad money, which is one of the most commonly used measures, is the most unstable. As Table 3 shows, the coefficient estimates for the effective broad money demand equation are the most precisely estimated, i.e., they have the narrowest confidence intervals, while the coefficient estimates for the ruble broad money demand equation are the least precise—in fact, and the coefficient on the deposit rate is insignificant for ruble broad money. Moreover, recursive estimates show that the confidence intervals for the coefficient estimates of the effective broad money demand function are stable, while they widen over time for the other money demand functions.²⁸

C. Short-Run Inflation Dynamics

To study the short-run dynamics of inflation, we combine the long-run inflation equation and the long-run money demand equation in an equilibrium-correction model. Since we found inflation and its long-run determinants to be I(1), we need to estimate the short-run equation in differences for OLS to be valid.²⁹ An equilibrium-correction model for the change in inflation is thus described by the lagged differences of its explanatory variables, an equilibrium correction term for inflation, and an equilibrium correction term for money. In addition, we add the lagged values of changes in money growth itself as possible determinants of short-run inflation.³⁰ This gives the following equation:

$\begin{array}{l}{\Delta }{\Delta }{p}_t=b_0+b_1{\Delta }{\Delta }{p}_{t-1}+b_2{\Delta }{\Delta }{p}_{t-2}+b_3{\Delta }{\Delta }\mathit{\text{ul}}{c}_{t-1}+b_4{\Delta }{\Delta }\mathit{\text{ul}}{c}_{t-2}+b_5{\Delta }{\Delta }{\mathit{\text{neer}}}_{t-1}+b_6{\Delta }{\Delta }{\mathit{\text{neer}}}_{t-2}\\ +b_7{\Delta }{\Delta }{p}_{\mathit{\text{ut}},t-1}+b_8{\Delta }{\Delta }{p}_{\mathit{\text{ut}},t-2}+b_9\mathit{\text{EC}}{\Delta }{p}_{t-1}+b_{10}{\mathit{\text{ECm}}}_{t-1}+{\displaystyle \sum _{i=1}^6}{b}_{10+i}{\Delta }{\Delta }{m}_{t-i},& & (6)\end{array}$

where EC∆p_t−1 indicates the equilibrium correction term (EC) for inflation (i.e., deviations from the long-run inflation equation), and ECm_t−1 indicates the equilibrium correction term for a given monetary aggregate (i.e., deviations from long-run money market equilibrium). The coefficient b₉ thus captures the short-run “correction” of inflation to temporary deviations from the long-run equilibrium in the goods market (where prices are determined), while the coefficient b₁₀ captures the short-run correction of inflation in response to temporary deviations from the long-run equilibrium in the money market. In addition, by including the terms ${\displaystyle \sum _{i=1}^6}{b}_{10+i}{\Delta }{\Delta }{m}_{t-i}$, we allow for a possible direct effect of an acceleration money growth on the acceleration in inflation.³¹

The estimated equilibrium-correction model suggests that the best model of short-run inflation is obtained when using effective broad money (EBM) as a monetary aggregate. Looking at the goodness-of-fit indicators in Table 4, the EBM equation has the lowest sigma, the highest R-squared, the maximum log-likelihood, and the smallest residual sum of squares (RSS). Looking at the residual tests, the EBM equation has the highest p-values, which implies that the EBM equation gives us the most certainty that the residuals are distributed normally and do not exhibit any significant autocorrelation or heteroskedasticity.

Table 4.

Estimated Short-Run Inflation Equations, January 1997–January 2004 ¹

Notes: *** indicates significance at the 1-percent level, ** at the 5-percent level, and * at the 10-percent level.

¹The regressions include dummies for 8-9/98, 1/99, 8-9/99, 1/00, and 2/00.

Table 4.

Estimated Short-Run Inflation Equations, January 1997–January 2004 ¹

	RCC	NRM	RBM	BM	EBM
∆∆pt-1	-0.186	-0.085	-0.112	-0.456 **	-0.544 ***
∆∆pt-2	0.219	0.323 *	0.341 *	0.425 **	0.354 **
Constant	0.047	0.096	0.055	-0.024	0.033
∆∆neert-1	0.350 ***	0.396 ***	0.379 ***	0.281 ***	0.165 ***
∆∆neert-2	-0.024	0.027	0.034	-0.012	-0.007
∆∆put,t-1	0.257	0.452 **	0.390 *	0.485 ***	0.394 ***
∆∆ put,t-2	0.155	0.031	0.039	0.030	0.038
∆∆ulct-1	0.044	0.085 *	0.088 *	0.000	0.008
∆∆ulct-2	0.118 ***	0.124 ***	0.148 ***	0.057	-0.002
EC∆pt-1	-0.168 **	-0.150 *	-0.138 *	-0.120 *	-0.085
ECmt-1	0.719	0.976	0.214	0.373	3.709 *
∆∆mt-1	-0.004	0.006	0.010	0.020	0.056
∆∆mt-2	-0.074	-0.104 *	-0.063	-0.290 ***	-0.259 ***
∆∆mt-3	0.140 **	0.098	0.172 *	0.285 ***	0.155 ***
∆∆mt-4	0.065	-0.063	-0.135	0.000	0.223 ***
∆∆mt-5	0.047	0.033	0.068	0.077	0.087 **
∆∆mt-6	-0.031	-0.069	-0.138 *	-0.026	-0.005
total effect of ∆∆m	0.144	-0.101	-0.085	0.067	0.255
No. of observations	85	85	85	85	85
No. of parameters	22	22	22	22	22
Sigma	1.56	1.59	1.59	1.34	1.13
R2	0.94	0.94	0.94	0.95	0.97
Log-likelihood	-145.85	-147.30	-147.27	-132.49	-118.61
RSS	153.93	159.28	159.18	112.42	81.09
F(21,63)	45.36 ***	43.73 ***	43.76 ***	63.21 ***	88.79 ***
DW	1.71	1.92	1.86	2.26	1.92
AR 1-6 test (F-test)	1.28 (0.28)	0.97 (0.46)	1.01 (0.43)	0.73 (0.63)	0.15 (0.99)
ARCH 1-6 test (F-test)	0.32 (0.93)	0.51 (0.80)	0.82 (0.56)	1.46 (0.21)	0.05 (1.00)
Normality test (χ 2-test)	1.05 (0.59)	0.57 (0.75)	0.76 (0.68)	1.96 (0.37)	0.69 (0.71)
Heterosk. test (F-test)	1.70 (0.08)	2.16 (0.02)	1.48 (0.15)	1.07 (0.44)	1.11 (0.40)
RESET test (F-test)	0.24 (0.63)	0.71 (0.40)	0.62 (0.43)	0.04 (0.83)	0.02 (0.88)

Notes: *** indicates significance at the 1-percent level, ** at the 5-percent level, and * at the 10-percent level.

¹The regressions include dummies for 8-9/98, 1/99, 8-9/99, 1/00, and 2/00.

View Table

Table 4.

Estimated Short-Run Inflation Equations, January 1997–January 2004 ¹

	RCC	NRM	RBM	BM	EBM
∆∆pt-1	-0.186	-0.085	-0.112	-0.456 **	-0.544 ***
∆∆pt-2	0.219	0.323 *	0.341 *	0.425 **	0.354 **
Constant	0.047	0.096	0.055	-0.024	0.033
∆∆neert-1	0.350 ***	0.396 ***	0.379 ***	0.281 ***	0.165 ***
∆∆neert-2	-0.024	0.027	0.034	-0.012	-0.007
∆∆put,t-1	0.257	0.452 **	0.390 *	0.485 ***	0.394 ***
∆∆ put,t-2	0.155	0.031	0.039	0.030	0.038
∆∆ulct-1	0.044	0.085 *	0.088 *	0.000	0.008
∆∆ulct-2	0.118 ***	0.124 ***	0.148 ***	0.057	-0.002
EC∆pt-1	-0.168 **	-0.150 *	-0.138 *	-0.120 *	-0.085
ECmt-1	0.719	0.976	0.214	0.373	3.709 *
∆∆mt-1	-0.004	0.006	0.010	0.020	0.056
∆∆mt-2	-0.074	-0.104 *	-0.063	-0.290 ***	-0.259 ***
∆∆mt-3	0.140 **	0.098	0.172 *	0.285 ***	0.155 ***
∆∆mt-4	0.065	-0.063	-0.135	0.000	0.223 ***
∆∆mt-5	0.047	0.033	0.068	0.077	0.087 **
∆∆mt-6	-0.031	-0.069	-0.138 *	-0.026	-0.005
total effect of ∆∆m	0.144	-0.101	-0.085	0.067	0.255
No. of observations	85	85	85	85	85
No. of parameters	22	22	22	22	22
Sigma	1.56	1.59	1.59	1.34	1.13
R2	0.94	0.94	0.94	0.95	0.97
Log-likelihood	-145.85	-147.30	-147.27	-132.49	-118.61
RSS	153.93	159.28	159.18	112.42	81.09
F(21,63)	45.36 ***	43.73 ***	43.76 ***	63.21 ***	88.79 ***
DW	1.71	1.92	1.86	2.26	1.92
AR 1-6 test (F-test)	1.28 (0.28)	0.97 (0.46)	1.01 (0.43)	0.73 (0.63)	0.15 (0.99)
ARCH 1-6 test (F-test)	0.32 (0.93)	0.51 (0.80)	0.82 (0.56)	1.46 (0.21)	0.05 (1.00)
Normality test (χ 2-test)	1.05 (0.59)	0.57 (0.75)	0.76 (0.68)	1.96 (0.37)	0.69 (0.71)
Heterosk. test (F-test)	1.70 (0.08)	2.16 (0.02)	1.48 (0.15)	1.07 (0.44)	1.11 (0.40)
RESET test (F-test)	0.24 (0.63)	0.71 (0.40)	0.62 (0.43)	0.04 (0.83)	0.02 (0.88)

Notes: *** indicates significance at the 1-percent level, ** at the 5-percent level, and * at the 10-percent level.

¹The regressions include dummies for 8-9/98, 1/99, 8-9/99, 1/00, and 2/00.

The estimation results also suggest that, in the short run, inflation does not significantly respond to excess supply of monetary aggregates that exclude foreign currency in circulation, but does significantly respond to excess supply of effective broad money. As Table 4 shows, the coefficient on lagged excess money (ECm_t−1) is insignificant in all cases, except for effective broad money. This may reflect the instability of the long-run equilibrium relationships for monetary aggregates that exclude foreign currency. The estimated coefficient for excess money in the EBM model suggests that excess money supply of one percent translates into a 3.7 percent acceleration in inflation (e.g., if inflation was initially 10 percent, it will be 10.37 percent the next month).³²

In addition, the results suggest that changes in EBM growth have the strongest and most persistent effect on short-run inflation. This is not just due to valuation effects (see Appendix II), since an acceleration in EBM growth still has a significant effect after accounting for the impact of exchange rate changes on inflation. While for most monetary aggregates, the effect of an acceleration in money growth occurs mainly after three months,³³ EBM growth continues to affect inflation even after four and five months, implying that its effect is more persistent. The total size of the impact is also largest for EBM, and equals 0.255, suggesting that the total effect of a one-percent acceleration in effective broad money growth is an acceleration of inflation by approximately ¼ percent. Perhaps intuitively, the second largest effect comes from ruble currency in circulation, with a total impact of 0.144, which is a bit more than half the impact of EBM. However, the total impact of changes in excess narrow or broad ruble money is negative, which again points to the instability of these money demand functions.

Finally, the estimates suggest that the speed of adjustment of inflation to its long-run equilibrium is about 6–12 months. As shown in Table 4, the coefficient on the lagged equilibrium correction term for inflation (EC∆p_1) varies from -0.168 to -0.085, depending on which monetary aggregate is used. (In the EBM equation, the coefficient equals -0.085 and is only significant at the 12 percent level). This means that, if inflation exceeds its long-run equilibrium by 1 percentage point, e.g., because of a temporary shock, 8.5 percent to 17 percent of this deviation is adjusted for every month, so that it takes about 6 to 12 months for inflation to return to its long-run equilibrium. Assuming that the EBM equation best describes the short-run inflation dynamics in Russia, this suggests that shocks to inflation are very persistent.

IV. Conclusion

Money demand in Russia, like that in other dollarized transition economies, has often been found to be highly unstable. In this paper, an estimate of foreign cash holdings outside the banking system was used to show that this instability can be explained in part by dollarization dynamics. More generally, our results suggest that inflation and money demand in dollarized economies cannot be well understood without taking foreign cash holdings into account.³⁴

We first estimated a long-run inflation equation using a markup model in which inflation is a weighted average of increases in unit input costs. We found that, in the long run, nominal effective depreciation accounts for roughly 50 percent of inflation, while the growth in unit labor costs accounts for about 40 percent and utility price growth for about 10 percent. We also tested the restriction that the marginal effects of input cost inflation add up to unity, and we could not reject the hypothesis that the inflation equation is, indeed, linearly homogenous, as theory suggests. The large pass-through from depreciation to inflation suggests that further nominal appreciation may help to reduce inflation.

We then estimated a set of long-run money demand equations for Russia using five different monetary aggregates, ranging from ruble currency in circulation to effective broad money, where the latter includes both foreign currency deposits and an estimate of foreign currency in circulation. We found that all measures of money demand that exclude foreign currency in circulation are strongly negatively dependent on the nominal depreciation rate, suggesting that foreign currency has been an important substitute for domestic money. The long-run depreciation semi-elasticity of ruble broad money demand was estimated at 0.4, implying that the appreciation of 2003 contributed about 5½ percentage points to the 30½ percent growth in real money demand. Moreover, we found that the money demand function for effective broad money, which includes foreign currency in circulation, was by far the most stable one and that money demand may appear to be unstable when foreign cash holdings are not taken into account.

Finally, we estimated an equilibrium-correction model for inflation in order to determine how the short-term dynamics of inflation are affected by deviations from the long-run inflation and long-run money demand equations. We found that the speed of adjustment of inflation to its long-run equilibrium is slow, and varies from 6 to 12 months. Inflation appears not to significantly respond to excess supplies of monetary aggregates that exclude foreign cash holdings, but does seem to respond significantly to an excess supply of effective broad money. In particular, we estimate that, for each 1 percent by which the effective broad money supply exceeds its demand, inflation accelerates by 3.7 percent.

Our estimates should be interpreted with some caution. First, our estimate of foreign cash holdings is subject to considerable uncertainty, and should be compared with other estimates (Appendix I). Second, our finding of a significant short-term impact of effective broad money on inflation may still be due in part to valuation effects (Appendix II), although we attempted to correct for the impact of exchange rate changes on inflation. Third, it may be difficult to interpret our estimated cointegrating vectors as true “long-run” equations, given the short time series and the structural break around August 1998, which may not have been entirely accounted for by crisis dummies.

Our estimated inflation and money demand equations can be further improved. With respect to the inflation equation, the unit labor cost estimates may need to be corrected for informal sector wages; Granger causality or weak exogeneity tests could be conducted to test for the direction of causality between wages and prices; and the robustness of the inflation results could be checked by using alternative measures of inflation (e.g., core or goods inflation) or other measures of utility costs (e.g., electricity tariffs). Possible ways to improve the money demand functions are to enhance the modeling of exchange rate expectations by, for example, including a ratchet variable to account for hysteresis (see Oomes, 2003; Kamin and Ericsson, 2003; and Mongardini and Mueller, 2000), and including measures of financial innovation and confidence in the banking system among money demand determinants.

We have several other suggestions for further research. Among other things, it would be useful to conduct an in-depth estimation of separate demand functions for each of the four components of effective broad money demand (ruble currency, foreign currency, ruble deposits, and foreign currency deposits) and study what determines the switches between these components, taking into account the relative returns on each component separately. In addition, the model could be extended by estimating additional long-run relationships for interest rates and exchange rates (where, ideally, all long-run relationships would need to be estimated simultaneously in a VAR). Finally, the usefulness of the model should be tested by constructing out-of-sample projections.

Keeping in mind these caveats, we conclude that estimates of foreign cash holdings can help us to better understand the relationship between money growth and inflation, and should therefore be taken into account when conducting monetary policy.³⁵ An acceleration in domestic money growth need not be inflationary to the extent that it reflects dedollarization and general monetization. Conversely, a slowdown in domestic money growth need not imply lower inflation to the extent that it reflects a slowdown in dedollarization or renewed dollarization.