Impact of monetary policy on the macroeconomy in Uzbekistan

This article intends to conduct a multifaceted empirical analysis of the relationship between the monetary policy of the Central Bank of Uzbekistan (CBU) and its policy goals (inflation, growth, etc.). At the end of 2019, the CBU changed the monetary policy operating system from the monetary targeting to the inflation targeting (IT) system and set a step-by-step inflation target after 2020 (10% a year until 2021 and 5% a year after 2023). However, as Uzbekistan’s economy is a small open economy, it is evaluated that the effectiveness of the interest rate policy transmission path is not high because of high economic volatility due to foreign factors and the undeveloped financial market. Related to this, empirical studies have shown controversial results about monetary policy performance after adopting inflation targeting in emerging markets (Aizenman et al., 2008; Cabral et al., 2016; Asab and Cuestas, 2021; Keefe, 2020).

In addition to general evidence on emerging markets, several studies emphasize that monetary transmission in small open economies often operates differently due to external vulnerability and shallow financial markets (Aizenman et al., 2008; Cabral et al., 2016). In transition economies, exchange rate regimes and institutional credibility play a decisive role in shaping policy effectiveness (Asab and Cuestas, 2021). For Uzbekistan, the IMF (2020) highlights the persistent influence of dual exchange rates and informal currency markets, which complicates the operation of the exchange rate channel. These contextual factors underscore the importance of explicitly considering exchange rate dynamics when assessing the effectiveness of monetary policy.

The channels through which monetary policy spreads to the real economy, such as prices and production, are typically the interest rate channel, exchange rate channel and credit channel. The first stage of the interest rate channel is a process in which monetary policy is sequentially spread to short-term market interest rates, long-term market rates and bank lending and loan interest rates within the financial market. The second stage is a process in which overall interest rate changes affect the real sector, such as consumption and investment. The exchange rate channel can be thought of as a process in which changes in domestic interest rates affect the exchange rate and a process in which fluctuating exchange rates affect the domestic real economy. The interest rate and the exchange rate channels are the processes in which monetary policy affects the price variables of the financial market and thus spreads to the real economy. On the other hand, the credit channel refers to the process by which monetary policy influences bank loans, which are quantitative variables, and spreads to the real economy. The spillover effect of such monetary policy may appear asymmetrically when easing and tightening monetary policies are implemented.

Against this background, the VARX model is constructed using the major monetary policy variables and domestic financial and real variables in Uzbekistan and the spillover effects of monetary policy through interest rates, exchange rates and credit channels are analyzed. Next, the asymmetric VARX model is established to analyze the asymmetric effect in order to compare the differentiated effects of changes in the monetary policy regime, such as tightening or easing. Both models consider the US dollar LIBOR interest rate and international oil price as exogenous variables to control for the influence of foreign factors. It can be said that it is essential to discover factors that restrict the effectiveness of monetary policy by comparing and analyzing the spillover effects of the CBU’s monetary policy on the macroeconomy through various empirical analyses.

Compared with previous studies on emerging markets and post-Soviet economies, this paper offers one of the first systematic empirical assessments of Uzbekistan’s monetary policy transmission, thereby filling a notable gap in the literature. By documenting both symmetric and asymmetric effects, our study not only extends the empirical scope of VARX-based analyses but also provides insights into the institutional specificities of a transitioning Central Asian economy.

A landmark study was carried out by Sims (1980), who pioneered the application of vector autoregression (VAR) model to economic fields. Since time series analyses frequently involve multiple variables, VAR model specification has been intensively employed in the empirical analyses, particularly in macroeconomics and finance. The VAR model has superiority over the conventional econometric approach (e.g. simultaneous equations) in the following aspects: (1) A statistical model is capable of capturing the dynamic relationship among multiple endogenous variables over time in that only lagged terms of endogenous variables and error term are incorporated in the equations. (2) The model is established based on statistical properties rather than strict economic theory. The only prior knowledge required is a list of variables which can be hypothesized to affect each other over time. (3) Albeit a large number of parameters to be estimated, part of which are inexplicable or lack economic significance, there are several analytical tools, such as generalized impulse response functions (GIRFs) and generalized (i.e. accounting for correlated shocks without orthogonalization) forecast error variance decomposition [GFEVD]).

Additionally, according to a body of existing literature, it is possible to take into account exogeneity in the system. Therefore, we employ a VAR model with exogenous variables as the baseline model specification (VARX), whose mathematical expression is as follows:

where Y_t is a k-dimensional vector of endogenous variables, whose pth-lagged vectors are denoted Y_t−i (i = 0,1, … p). X_t−i (i = 0,1, … r) represents a d-dimensional vector of exogenous variables. A_t and B_t are k*k endogenous coefficient matrices and k*d exogenous coefficient matrices, respectively. $εt$ indicates a k-vector of error terms, which must satisfy three conditions as follows: (1) Every error term has a mean of zero. (2) The contemporaneous covariance matrix of error terms is a k*k positive-semidefinite matrix denoted Ω. (3) There is no correlation across time. In particular, there is no serial correlation in individual error terms.

In economics, VARX models have become popular to model small open economies which are potentially exposed to external shocks, especially from influential developed countries, such as the USA. In addition to endogenous variables related to transmission channels of monetary policy, we incorporate two exogenous variables in the system, namely, WTI oil price and the US three-month LIBOR interest rate.

In this study, we employ money aggregate (M0 and M2) to proxy for Uzbekistan’s monetary policy, owing to the inaccessibility of the interbank offer rate and lack of statistical significance of the change rate in the domestic policy rate. With respect to the interest rate channel, the key factor is proxied by the long-term market interest rate (average rate of deposit and lending rates over one year). The nominal exchange rate of the US dollar against the Uzbekistan som and the amount of loans are key variables in the exchange rate channel and credit channel, respectively. Other endogenous variables consist of the domestic inflation rate and real gross domestic product (GDP) growth since we aim at analyzing the impact of monetary policy on macroeconomic variables.

Money aggregate and real GDP series are seasonally adjusted before calculating growth rates. Besides, we convert real GDP series from quarterly to monthly by using the quadratic match sum method, which is a standard temporal disaggregation technique that distributes low-frequency data into higher-frequency series while preserving consistency with the original aggregates (Denton, 1971; Chow and Lin, 1971). Market interest rates and the US three-month LIBOR series are transformed by first-order differencing. CPI is seasonally adjusted based on the month-on-month inflation rate. Regarding the calculation of growth rate, instead of taking the log-difference, we firstly take the current value and subtract that from the previous value. Then, this difference is divided by the previous value and multiplied by 100 to get a percentage representation of the growth rate. All series span the period from 2013M2 through 2021M3 in monthly frequency. Variable description and data source are presented in Table 1.

In the Uzbek context, the Central Bank historically relied on monetary aggregates as its main operational target, particularly under the monetary targeting regime that prevailed until the end of 2019. Even after the transition to inflation targeting, liquidity management continued to be implemented through monitoring and controlling reserve money and broad money growth, reflecting the underdevelopment of short-term financial markets and limited transmission of the policy rate. Therefore, M0 and M2 serve as valid proxies for monetary policy stance as fluctuations in these aggregates effectively capture changes in liquidity conditions induced by the central bank.

Table 2 reports the summary statistics of data with an outlier (2017M9) removed. Regarding the unit root tests, we conduct both the augmented Dickey–Fuller (ADF) test and the Phillips–Perron (PP) test. The results (see Table 3) show that all series are stationary at the 1% significance level. (The PP test for GDP is significant at 10%.) Additionally, the lag length tests are performed to select the appropriate lag length. To prevent from losing more information and preserve a higher degree of freedom (DOF), the model specification with one lag length (p = 1) is selected based on the Bayesian information criterion (BIC), which penalizes model complexity more heavily.

In addition to the baseline model, we investigate the asymmetric spillover effects of monetary policy of the CBU on the real economy by using the asymmetric VARX model specification, which is used to analyze the asymmetry of tightening or easing monetary policy. Money aggregate series fall into two categories representing expansionary monetary policy with non-negative values (all negative values are replaced with zero) and contractionary monetary policy with non-positive values (all positive values are replaced with zero). The asymmetric VARX can be written as follows:

where Y_t is the vector of endogenous variables, which are identical to Equation (1). X_t denotes the vector of exogenous variables. MP_t,tight and MP_t,ease represent the tightening monetary policy and easing monetary policy, respectively. In the empirical analyses, we conduct the GIRFs and GFEVD based on estimation results of VARX and asymmetric VARX. The IRFs map out the response of the system over time to a specific shock pattern, for instance, 1-SD shock in this study. GFEVD decomposes the variance of the forecast error into the contributions from a specific shock at a given horizon in the form of a percentage.

Figure 1 shows the impulse responses of all endogenous variables to a 1-SD shock to M0 in the baseline model, with dotted lines representing ±2 standard errors confidence bands. Specifically, a 1-SD shock to M0 is associated with a statistically significant decline in the long-term interest rate at the initial stage. Meanwhile, a positive shock to M0 results in an increase in the nominal exchange rate of USD against UZS and the inflation rate, both of which reach their maximum effect in the first month. Although an unexpected expansionary monetary policy causes the amount of loan and output to rise, the effects are not statistically significant over the time horizon.

Figure 2 through Figure 3 reveal the responses of CPI and GDP to a positive shock to long-term interest rate, nominal exchange rate and credit growth, respectively. There is little empirical evidence to support the existence of the interest rate channel of monetary policy in Uzbekistan. However, the monetary policy transmission through the adjustment of the exchange rate can be concluded. In general, the changes in the central bank’s policy rate exert influence on various commercial interest rates, such as overnight call rates, deposit and lending rates. If the central bank raises its base rate, short-term interest rates rise immediately, and then mid- and long-term interest rates are likely to be on an underlying upward tendency. Such changes in the interest rates are bound to affect aggregate demand in terms of both investment and consumption. For instance, the monetary authority carries out an expansionary monetary policy that seeks to stimulate the economy by increasing the money supply, which results in a decrease in the interest rates and an increase in aggregate demand. Nonetheless, as a consequence of adopting monetary aggregates as the intermediate target by 2019, the effectiveness of monetary policy through the interest rate channel is limited (see Figure 4).

On the other hand, in Uzbekistan, a small open economy, exchange rate fluctuations due to external shocks have relatively significant effects on the domestic price level and real economy. Moreover, the impact of domestic credit growth shock on real GDP growth is not statistically significant. In contrast, the inflation rate, to a greater or lesser extent, responds positively to domestic currency depreciation and a credit growth. However, the exchange rate change and credit growth are not effective in stimulating the economic growth. Expansionary monetary policy will provide more liquidity in the supply of credit, which subsequently stimulates credit-based consumption by households and investment by businesses raising funds through bank loans.

Tables 4 and 5 present the results of generalized forecast error variance decomposition (GFEVD) of inflation and real GDP growth. To evaluate the relative importance of other variables, forecast error variance decomposition is performed and the contribution of other variables to each endogenous variable is compared for each forecast horizon. This variance decomposition is to understand how much the variance of forecast error for each endogenous variable is explained by the variance of itself and other variables. The contribution of inflation to the variation in itself still accounts for more than 60% even after one year, while the contribution from the exchange rate was 20% at the beginning, relatively high compared to other variables. Both the contribution of the long-term interest rate change and domestic credit growth to the variance decomposition of inflation are less than 10%. Regarding the variance decomposition of the real GDP growth, the contribution of the exchange rate change after one year accounts for 13.8%, which is relatively high compared to the contributions from the long-term interest rate change (6.1%) and domestic credit growth (0.2%).

Comparing the reactions of endogenous variables to a shock in M0, M2 is no longer an effective determinant factor in controlling inflation and long-term interest rate, but it significantly boosts the credit growth (see Figure 5). Meanwhile, our findings are quite robust since the impulse responses of price level and GDP growth under three main channels are similar to that of the VARX model with M0 in terms of signs and significance (see Figures 6–8). Specifically, with the unanticipated hike in the long-term interest rate, the impulse response functions of the inflation rate increase by 0.05 SD immediately, followed by a peak in the second month. The economic growth initially declines before moving towards the horizontal line, but the statistical insignificance suggests that the interest rate channel is ineffective.

Following a rise in the nominal exchange rate, the price level rises, then gradually dying out as the impact returns to almost zero within a few months. And the real GDP growth is negatively responded to domestic currency depreciation. A spike in price level occurs at the time of the credit growing, and the impact converges back to zero after half a year, while the economic growth appears irrespective of a shock in loans, implying that the credit channel is fairly weak as a monetary policy transmission.

Table 6 presents that the price level is primarily influenced by itself, accounting for more than 60% of GFEVD over the forecast horizon. The contribution of long-term interest rate and domestic credit growth to variations in the CPI remains stable at a lower level, roughly 4 and 9% in one year, respectively. The analogous value for exchange rate contributes 21% in the initial period, followed by a slight drop and leveling off at approximately 17%. These results indicate that the exchange rate dominates over other factors in regulating inflation.

At the outset, roughly 91% of the variations in the GDP are from shocks to the GDP itself and the remaining 9% are from other five endogenous factors. The system becomes stable after six months, with the shocks in interest rate, exchange rate and domestic credit contributing to GFEVD in GDP about 6, 11 and 0%, respectively (see Table 7).

Figure 9 shows the impulse response functions for an asymmetric VARX model with a positive series of money supply rather than the original series. Results show the long-term interest rate is negatively responded to easing monetary policy, which is widely consistent with monetary theory. The impulse responses of the nominal exchange rate change, credit growth and price level to a 1-SD shock to expansionary monetary policy are very similar to that of the baseline model. However, the output shows a very weak significant response to expansionary monetary policy over the short run.

Regarding the interest rate channel, an unexpected shock in the long-term interest rate produces positive innovation in the consumer price. Meanwhile, real GDP growth drops slightly before reaching a trough, and an increase is recorded afterward. The inflation continually rises until it stagnates at the benchmark level as a result of depreciation of UZS. However, the real GDP growth falls in response to a positive shock to the exchange rate. In analyzing the credit channel, the positive reaction of the price level to credit growth remains statistically significant for less than two months. The output shows no significant response to credit growth over the time horizon (see Figure 10 through Figure 11) (see Figure 12).

According to Tables 8 and 9, approximately 60% of variation in CPI is contributed by the price level itself, followed by exchange rate change (17%), credit growth (9%), GDP (6%), long-term interest rate (4%) and positive M0 growth (3%) in one year. Within the same forecast horizon, the contribution of exchange rate to the GFEVD in GDP plays a leading role in comparison with the other two channels.

Figure 13 presents the impulse responses of variables to a negative M0 growth shock. Not surprisingly, the contractionary monetary policy results in a fall in long-term interest rate and real GDP growth, which is declined by 0.3 SD before reverting very slowly back to the baseline. However, none of the factors are statistically significantly responded to external shocks, due partly to insufficient observations that the CBU has been adopting easing monetary policy for a long period of time. With respect to the IRFs under three channels, there is no significant difference with that of the positive M0 growth equation (see Figures 14–16). It should be noted that in the contractionary specification, the M0 series is defined by non-positive values. Therefore, a positive innovation to M0_NEG effectively represents a contractionary monetary policy shock, with the sign of the response interpreted in the opposite direction compared to the expansionary case.

Meanwhile, the contribution of the exchange rate change to the variance decomposition of inflation is 16.9% after one year, which is higher than that of the domestic credit growth (9.4%) and long-term interest rate (4.1%). In the variance decomposition of the real GDP growth rate, the contribution of the exchange rate change rises from 1.5% in one month to 10.6% one year later, while the contribution of the long-term interest rate after one year is 5.4% and the contribution of the domestic credit growth rate is 0.2%, which is relatively less explanatory (see Tables 10 and 11).

This pattern reflects the broader policy regime context in Uzbekistan, where contractionary measures were rarely employed. The authorities maintained a strong bias toward easing policies to support growth and financial sector development, making contraction politically less feasible and limiting observable variation in contractionary shocks during the sample period.

In this article, we have analyzed the spillover effects of monetary policy on macroeconomic variables related to Uzbekistan’s economy by employing the VARX approach. Furthermore, using the asymmetric VARX model specification, we attempt to identify the asymmetric spillover effects of expansionary and contractionary monetary policies over the sample period from 2013M2 through 2021M3.

The impulse response functions suggest that both exchange rate and price level soar up statistically significantly in response to a shock to money supply, whose increases have little impact on domestic credit and real GDP growth. In stimulating economic growth and managing inflation, the exchange rate channel is effective, while the credit channel is only valid for inflationary pressure. However, we find no evidence of an interest rate channel over the sample period. In addition, the results of generalized forecast error variance decomposition provide evidence that the contribution of exchange rate change in inflation rate and real GDP growth is relatively higher than that of domestic credit growth and long-term interest rate.

Meanwhile, with respect to IRFs based on asymmetric VARX, a positive shock in M0 growth leads to a fall in the long-term interest rate and depreciation of domestic currency. The inflation and real GDP growth statistically significantly rise in the early periods. Regarding the exchange rate channel, an increase in the nominal exchange rate results in a hike in price level and a drop in real GDP growth. While following an increase in domestic credit, inflationary pressure grows through the credit channel. The impulse responses of real and financial variables to a negative shock in M0 growth are not statistically significant. The exchange rate plays a relatively important role compared to long-term interest rate and domestic credit growth in terms of GFEVD.

In sum, regarding the ineffectiveness of the interest rate channel, the monetary authority needs to improve financial market development and liberalization, for example, revitalizing the short-term financial market. Also, the CBU should continuously monitor conditions in the foreign exchange market since the exchange rate channel is the most effective mechanism through which monetary policy is transmitted to the real sector in Uzbekistan. Meanwhile, domestic credit growth does not have a positive effect on real GDP and acts as a domestic inflation pressure, which means that the CBU should properly control the increase in bank credit as necessary. In this regard, it is necessary to establish the integrated policy framework (IPF) that jointly considers monetary policy, foreign exchange policy and macro-prudential policy (IMF, 2020). The IPF helps enhance monetary autonomy, improve price stability and reduce output volatility. Relatedly, monetary policy and macro-prudential policy are complementary from the perspective of an integrated inflation targeting that the central bank wants to achieve and maintain price and financial stability (Agénor and da Silva, 2019). Therefore, the central bank should consider credit growth in setting policy interest rates and calibrate monetary policy jointly with macro-prudential tools.

In addition, more concrete measures are needed to strengthen the monetary transmission mechanism. The authorities should foster the development of a short-term money market to enhance the interest rate channel, improve transparency in exchange rate management to strengthen policy credibility and gradually reduce fragmentation between official and parallel markets to mitigate distortions in monetary transmission.

One limitation of this study is that the analysis relies on the official exchange rate, while Uzbekistan maintained a dual exchange rate regime during much of the sample period. The potential discrepancy between official and parallel market rates may bias the estimated effects of exchange rate movements on inflation and output. Future research could consider incorporating a distortion indicator, such as the spread between official and black-market rates, when such data are available.