Literature Review: Monetary Policy Transmission Mechanism

We start the chapter with reviewing the empirical literature related to the monetary policy transmission mechanisms, as previous studies provide us with necessary tools and ideas for performing the empirical analysis. Based on the literature reviewed, we choose the appropriate model for our analysis. The next section is devoted to describing the data employed. The final part of the chapter addresses the questions regarding the empirical estimation and interpretation of obtained results.

4.1 Empirical literature review

In our empirical literature part we focus mainly on the studies conducted in regard to transition and emerging economies. The cases of developed countries are discussed to give the general idea about the variety of results between transition and developed country cases reached in empirical analysis of transmission mechanisms. We did not include all research examined by us regarding the monetary transmission, but those which can be seen as representatives to other similar studies.

Mark Gertler and Simon Gilchrist (1993)

This paper was one of the first analyses conducted regarding the role of credit market imperfections in monetary transmission, when debate on this topic was heating. The authors analysed the responses of different forms of credit as well as different type of borrowers to the monetary shocks and found two ways of relevance of credit market imperfections to the transmission mechanisms: (1) certain type of borrowers (mainly small firms) may be forced to rely on bank credit as a result of credit market imperfections; (2) the same type of borrowers may become excessively sensitive to movements in interest rates (the authors used a term “excess sensitivity hypothesis”, in which frictions in the credit market may amplify the distortions in borrowers’ spending decision, that can offer explanation for the second point).

Moreover, by conducting an empirical analysis they showed sharp differences in behaviour of large and small firms. Following the monetary tightening, bank loans to small firms declined while bank loans to large firms increased. The offered explanations suggested that large firms try to smooth the impact of declining sales by using bank loans or accessing short term credit markets, whereas small firms may not have access to these facilities, even though they might suffer large drops in sales, which leaves small firms with the only option of cutting the production. So, all again comes to credit market frictions, where imperfect information may lead to these types of outcomes.

Ben S. Bernanke and Mark Gertler (1995)

In this well cited paper the authors argued that the transmission mechanism had been treated as a “black box” by empirical studies, as there were shortcomings of explanations. For instance, the authors provided counterarguments regarding the magnitude of effects on spending caused by interest rate shock, mentioning the common finding that variables such as lagged output or cash flows have larger effect on spending.

Moreover, in order to examine how the traditional monetary transmission facts are hold, such as the directions of output and price movements, as well as their lagging periods, the authors undertook an empirical analysis, employing VAR method, in which a federal funds rate was used as an indicator of the stance of monetary policy. The results obtained were in line with conventional facts, namely, the decline in GDP as a response to monetary tightening begins after around four months later, whereas the price decline is lagged around a year.

The authors reached a conclusion, saying that as a part of credit channel, the role of the bank lending channel had been diminishing over time, mainly due to the financial innovations and deregulation. Similar conclusion was reached by Mishkin (1996) regarding the bank lending channel in the United States, due to the changes in the regulatory framework and changes in traditional bank lending business. At this point we have to note that this is not true for Uzbekistan, where capital markets are in very early stage of development and banks are still the main player in the intermediation process.

Another point made by the authors that is notable to mention, though not fully related to our analysis, goes to the forecasting power of credit aggregates. The authors warned against treating credit aggregates as an independent causal factor affecting the economy, arguing that credit is not a primary driving force, but the credit conditions are (e.g., measure by the external finance premium).

Veronica Babich (2001)

In this paper the author examines monetary policy mechanism in Latvia for the sample period of 1992–2000, employing narrative approach instead of standard VAR method, which makes it worthy of a mentioning.

In order to investigate different monetary channels the author puts forward several hypotheses regarding these channels and tests them with different means. For example, the author looks tests interest rate hypothesis by looking at presence of “abnormal” increases in interest rates following the monetary shocks. The findings suggest that transmission is working through interest rate and credit channels, but the effects on output and prices seems to be weak or even non-existent.

Igor Vetlov (2001)

This paper is another study employing narrative approach for studying monetary transmission mechanisms for Lithuania, which is one of the Baltic countries with a currency board arrangement (CBA). This aspect of monetary policy has its important implications for the monetary transmission, as the limitation of policy may change the relative importance of channels.

The author finds weak transmission of developments on money market to bank interest rates and explains this finding by lack of competition in the financial markets of the country. As a suggestion to improve this channel the author proposes closer integration with international markets. Another finding is that domestic investment and private consumption is weakly affected by the changes in the interest rates, mostly because less dependency on bank loans for financing the expenditures. Regarding the exchange rate channel, the author finds significant transmission, based on fact that the inflation declined from 5.1 percent in 1998 to 1 percent in 2000 due to the exchange rate shocks.

Another important aspect of this paper is possibly the examination of expectations channel, which is omitted due to the lack of data. The author explains developments in this channel and points out that the Lithuanian CBA has experienced periods of low level credibility, though the CBA usually expected to increase credibility, thus to reduce inflationary expectations.

Georgy Ganev, Krisztina Molnar, Krzysztof Rybinski, Przemyslaw Wozniak (2002)

The authors examine interest rate and exchange rate channels of monetary transmission in ten Central and Eastern Europe countries (CEEC) using impulse response analysis for the sample period 1999–2000. In addition to the empirical analysis the authors make a good effort to review and summarize large number of empirical literature regarding the monetary transmission in CEEC and highlight that the literature does not hold much evidence regarding the clear monetary transmission channels. For instance, while weak first step transmission is more common (though links between changes in policy variables and certain credit or monetary aggregates are rare), the existence of a significant second step transmission is generally not found.

The findings confirmed previous empirical studies regarding the importance of the transmission channels considered. In particular, the authors find more stable exchange rate channel in most of the countries. The interest rate channel was stable only in Slovenia during the whole sample period and significant in Poland and Czech Republic for most sub-samples, whereas was insignificant for Hungary, Latvia, Romania. It seems Slovakia is the most lagging country, as the exchange rate channel becomes significant at the end of the period and interest rate channel remains insignificant during the entire sample period.

Ignazio Angeloni, Michael Ehrmann (2003)

This paper examines the interest rate, banking and stock market channels of transmission in euro area after 1999 and attempts to find out whether the change in transmission mechanisms after European Monetary Union (EMU) has occurred or not.

The authors use monthly data on lending and deposit interest rates for five euro area countries and find that after around 1999 the euro area banks reveal stronger and increasingly homogeneous response of bank rates to the monetary authority signals. The authors conclude that this change is due to monetary union, as this effect is not found non-euro area countries.

Regarding the interest rate channel, the authors conclude that it has changed even before EMU and by now affects euro area economies largely in a similar way.

The authors use ten national stock indices and estimate the effect of policy change by the European Central Bank on these national indices. The findings are as expected: following the monetary tightening stock indices decrease in all countries, except Ireland, where the effect is positive, though insignificant. In this part the authors do not make comparisons with pre-EMU situation, which makes it impossible to differentiate the effect of EMU on the stock price channel.

Jesus Crespo-Cuaresma, et.al. (2004)

In this paper the authors employ autoregressive distributed lags (ARDL) model in order to analyse the features of the interest rate pass-through in the Czech Republic, Hungary and Poland. The findings are significantly different not only across countries, but also across market interest rates. For instance, the interest rate pass-through turns out to be complete for all market interest rates in case of Poland and for only some market rates in Hungary, whereas the pass-through is incomplete in the Czech Republic (with exception of the interbank rates). The authors conclude that the findings are in line with the view that the interest rate pass-through should increase with the longer usage of the direct inflation targeting, as price stability starts becoming common.

Jerome Hericourt (2006)

Hericourt conducts an empirical analysis regarding the monetary policy transmission in the eight Central and Eastern European countries (CEECs), newly joined the EU. In order to analyse the interest rate, the exchange rate and the domestic credit channels the author estimates different unrestricted VAR models for each country. The findings are mostly similar with those for other transition countries. One aspect of this study is that the author undertakes close look at credit variance and finds increasing interest rate share that explains variation and concludes that it may support the hypothesis of an evolving credit channel in majority of the countries under consideration.

Departing from the previous studies, the author shows that the importance of the exchange rate channel is decreasing comparatively to the interest rate channel (except Estonia and Latvia), concluding that almost all the countries under study are converging towards the old euro area members, which gives optimistic view for close integration.

Era Dabla-Norris and Holger Floerkemeier (2006)

In this paper the authors provide an evaluation of monetary transmission mechanisms in Armenia, analyzing interest rate, exchange rate and credit channel by estimating unrestricted VAR model. The findings are similar with those for other transition economies. For instance, the rapid exchange rate pass-through (within 2 months) and only modest impact of interest rate changes on prices suggest that exchange rate plays an important role in monetary policy of Armenia. Regarding the credit channel, the authors find that a shock to bank loans leads to nearly instant increase in price level that keeps significant for four months, while output does not yield significant response.

As a conclusion, the authors suggest several ways that can help to increase the effectiveness of the transmission channels, which are generally true for many transition economies, including Uzbekistan. These suggestions include reducing the size of the shadow economy through due regulation and supervision, effectively removing the excess liquidity by various market operations and increasing the efficiency and transparency of banks.

S. Drobyshevsky, P. Trunin, M. Kamenskikh (2008)

In this paper the authors investigate monetary transmission mechanisms in Russian economy for the sample period of 1999–2007, by employing VAR approach and an alternative approach by using microdata to closely analyse the credit channel.

The authors find that during the sample period the shocks in policy variables had no affect on output and inflation, which implies absence of transmission mechanisms. These results are explained by several factors, such as the changes after the Russian crisis may not be captured by the variables and lack of long data series.

In the alternative approach the authors use quarterly data from the balance sheets of banks for the period of 2000Q1–2007Q3 and reveal that the increase of interest rates in the interbank market leads tot the decrease in loans to households and businesses. Moreover, the authors find that the level of decrease depends on the size of the businesses and the adequacy of capital. We can see that these results are similar with those found in Gertler and Gilchrist (1993) that we have reviewed above in the text.

Charalambos Tsangarides (2010)

This is one of the very recent papers regarding monetary transmission channels, in which the case of Mauritius was investigated, using monthly data for 1999–2009, employing both structural and unrestricted VAR methods. Unrestricted VAR results show that a policy shock (increase in repo rate) has modest effects both on output (0.5 percent decline) and prices (0.2 percent decline), which appear after 4 months of the shock. The inflation showed to be more prone to the shock, keeping lower than pre-shock level for more than 10 months, while output returned to its pre-shock in 6 months. With structural identification model the authors find that the repo shock transmits almost immediately after the shock, while the shock to the exchange rate is similar to the unrestricted model case.

Another notable aspect of this study, in our opinion, is that the models are estimated for headline and core CPI separately, which may give different insight to the transmission mechanisms. For instance, the authors find that transmission of shocks to the monetary aggregates to the output and prices is stronger in case of core CPI modelling compared to the case of headline CPI.

The authors conclude that their results are typical for transition economies, which is described by stronger transmission for nominal variables, such as inflation rather than real variables, such as output.

Jean Boivin, Michael T. Kiley, Frederic S. Mishkin (2010)

This research analyses the evolution of monetary transmission over time, examining the changes in transmission mechanisms caused by developments in the financial markets and in the monetary policy. Empirical research is undertaken employing VAR approach by using quarterly data for the United States for quite a long period of time – from 1966 to 2008.

The authors find that the magnitude of responses of GDP is similar over time, though the response was much faster and short-lived until 1979. Another interesting point of the VAR is related to the change in importance of the policy shock over time. For instance, relative to its standard deviation, in the 1979–1984 period policy shock was 1.5 times smaller than pre-1979 period.

In the alternative approach the authors expand the VAR by adding disaggregated variables, such as consumption (durable goods and services) and investment (residential and non-residential). Interestingly, when these variables are added the estimations were different from the previous case. For instance, the response of the GDP was changed as the monetary policy shock on consumption and investment was reduced more than benchmark VAR estimation.

Notable part of this study is that the authors empirically show that the VAR conclusions are largely dependent to the choice of variables included, which is necessary in our empirical analysis, where we attempt to justify our choice of variables. Our choice might be far from perfect, leaving room for further experiments to find variables that fits models better.

4.2 Model specification

Based on our literature review, we can distinguish several approaches to the study of monetary transmission mechanisms.

One possible approach is based on descriptive and comparative analyses, as described by Ganev et al. (2002), also known as a “narrative method”. This approach starts with identifying shocks in the monetary environment, then developing a counterfactual, and finally drawing conclusions based on comparing the actual with the counterfactual. Though not frequently used, this method is adopted by several authors, such as Babich (2001) for the case of Latvia and Vetlov (2001) for Lithuania.

Another method is estimation of various econometrics models. Ganev et al. (2002) provides three types of models that can be used in this approach:

Using reduced form VAR models, that attempt to avoid strong constraints on the data.

Using small structural models, which consider specific aspects of transmission mechanisms.

And finally, using large macroeconomic models with several equations that try to capture various links in the monetary transmission.

The third approach is described by analyses using micro data that attempt to explain changes in behaviour of agents to the changes in the monetary policy. This approach is used by Drobyshevsky et al. (2008) for Russia.

Probably, the VAR estimation is the most frequently used approach among researchers, which was originally pioneered by Sims (1980), as overwhelming part of the literature is based on this method. Sims (1980) showed that reduced form unrestricted [1] (only restrictions are on the lag length and the choice of variables) models can be used for policy analysis, arguing that many false restrictions were posed in existing models and these models are “nominally over-identified” [2] . The VAR approach has been employed in studies regarding the monetary transmission not only in transition economies, but also in developed economies, such as Christiano et al. (2000) for the U.S. and Mojon and Peersman (2003) for the EU countries, which, according to Hericourt (2006), makes them even more appropriate for the transition economies.

So, when it comes to the question of choosing structural or reduced form VAR model, that will be adopted in our empirical analysis, we need to state advantages and disadvantages associated with these models.

Several authors state that unrestricted VARs are useful to investigate the monetary transmission mechanisms in the context of transition and emerging economies, where data series are usually short, continuous structural and institutional changes may put difficulties in the use of structural models, and where the economy has non-neoclassical characteristics (Dabla-Norris, Floerkemeier, 2006; Tsangarides, 2010; Ganev et al., 2002)

Another notable aspect of the reduced form VAR models is that they impose minimal restrictions on how monetary policy shocks affect the real economy, which can lead to more straightforward estimation (Dabla-Norris, Floerkemeier, 2006; Tsangarides, 2010). On the other hand, these minimal restrictions may have problems in capturing the institutional changes (Ganev et al., 2002). Furthermore, Boivin et al. (2010) state that analyses based on relatively unrestricted VAR models suffer from “the curse of dimensionality” [3] and have not reached uniform conclusion regarding the links between monetary policy and economic activity. In addition, these authors empirically show that unrestricted VAR conclusions are largely dependent on the choice of variables included in the VAR.

Regarding the drawback of minimal restrictions, Sims (1980) states that “…a more systematic approach to imposing restrictions could lead to capture of empirical regularities which remain hidden to the standard procedures and hence lead to improved forecasts and policy projections” [4] . But we have to admit that data necessary for estimation of these models can be unavailable or unreliable (Ganev et al., 2002).

Additional superiority of unrestricted VARs can be seen in the explicit recognition of simultaneous behaviour of monetary policy shocks and macroeconomic developments (Dabla-Norris, Floerkemeier, 2006; Tsangarides, 2010).

Based on our comparisons, we decided to adopt reduced form unrestricted VAR model in our analysis, which seems more plausible in context of the Uzbek economy, where both transition and data constraints are exist.

Following Dabla-Norris and Floerkemeier (2006), our model takes the following form:

,

where is vector of endogenous variables and is vector of exogenous variables and is vector of independently and identically distributed (i.i.d.) shocks. and are matrix polynomials in the lag operator L (matrices of coefficients to be estimated).

The endogenous variables include real GDP (rgdp), CPI (inf), refinancing rate of the Central Bank of Uzbekistan (refcb_uz), real exchange rate (rer), broad money (rm2), credit to the economy (credecon). Following Dabla-Norris and Floerkemeier (2006) the latter two variables are alternatively (one at a time) used in the model in order to capture the role of money and credit stock development in the monetary policy.

The exogenous set of variables consists of the Russian GDP (rus_gdp) and the federal funds rate (fed_rate). We have to mention that treating them as exogenous related to the small open economy hypothesis, where it is assumed that the economy is small and it can have only a negligible or no effect on the big countries.

Using these identifications, we can write the VAR in the following form:

The ordering of variables is as above, which follows Dabla-Norris and Floerkemeier (2006) and Hericourt (2006). The output is ordered before prices, assuming that it adjusts more slowly than prices. And monetary variables are ordered after output and prices, as they do not have simultaneous affect on the real sector variables, because of sluggish reaction of output and prices. This is another implicit assumption related to the “price stickiness” in the short run. The refinancing rate (nominal interest rate) responds contemporaneously to the changes in real variables (output and inflation), but not to shocks to the subsequent variables. The ordering of other policy variables follows the same logic. Lastly, the exchange rate is assumed to respond to all types of shocks.

Another implicit assumption that is worthy of mentioning is that in all empirical analyses it is presumed that the monetary authority (Central Bank of Uzbekistan, in our case) has all a full control over its policy tools, meaning that the authority can freely manipulate these tools.

4.3 Data description

The quarterly data covering the period of 1998:Q1 to 2009:Q4 is used in this study with a total of 48 observations (Figure##). We have to note that the GDP, CPI, monetary aggregates and domestic credit data are non-publicly available, but working data, extracted from the Central Bank of Uzbekistan. The questions regarding the accuracy of the data are addressed by comparing them with those published in the IMF country reports for the several previous years. [5] The data regarding the refinancing rate of the Central Bank of Uzbekistan and the UZS/USD nominal exchange rate is available on the web-site of the Central Bank. Finally, the data for the U.S. CPI, Federal funds rate and the Russian GDP is extracted from the IFS online database.

Figure ##. Data used for the empirical analysis (raw data)

Source: Author’s estimates.

Several of abovementioned variables are seasonally adjusted using ARIMA X12 process, before running the model, taking into account the results of F-tests for seasonality. The descriptive statistics after the adjustments are shown in Table ##.

Table ##

 Minimum

 Mean

 Median

 Maximum

 Std. Dev.

Observations

RGDP_SA

3,014.1

12,055.0

9,696.1

31,627.2

7,977.2

46

INF_SA

-3.2

3.3

2.4

8.7

2.7

46

REFCBUZ

14.0

21.8

20.0

36.0

8.0

46

RM2_SA

1,860.0

6,407.7

3,663.4

20,093.4

5,303.9

46

CREDECON_SA

260.8

2,820.8

2,890.9

7,691.4

1,917.1

46

RER_SA

88.9

2,136.5

2,420.5

4,549.6

1,395.0

46

RUSGDP_SA

59.9

91.9

91.1

123.9

19.2

46

FEDR_SA

0.1

3.3

3.4

6.6

2.0

46

As in any multi-variable empirical analysis we start with testing for the level of integration. In our study we use traditional ADF (Augmented Dickey-Fuller) Unit root test, with both SIC and AIC information criteria.

Table 1. Unit root test results

Variables

SIC

AIC

DLOGRGDP

-12.19336***

(0.0000)

-12.19336***

(0.0000)

INF_SA

-9.070677***

(0.0000)

-3.288186***

(0.0016)

REFCB_UZ

-3.704190***

(0.0077)

-3.704190***

(0.0077)

DRM2_SA

-5.516529***

(0.0002)

-5.516529***

(0.0002)

DRER_SA

-1.984511**

(0.0462)

-1.984511**

(0.0462)

D2CREDECON_SA

-8.687519***

(0.0000)

-8.687519***

(0.0000)

RUS_GDP_SA

-4.616015***

(0.0000)

-4.616015***

(0.0000)

DFED_RATE_SA

-2.827556***

(0.0057)

-2.827556***

(0.0057)

Note: *=10%, **=5%, ***=1% significance level; p-values are given in brackets, which are MacKinnon one-sided p-values.

Source: Author’s estimates

Detailed description of data is given below.

1. Real GDP

The real GDP is obtained by adjusting the nominal GDP by the CPI of the respective periods. Even though there might be significant differences between GDP deflator and CPI, the usage of CPI can be a proxy for obtaining real values. [6] Moreover, we couldn’t obtain data for the GDP deflator, making the usage of CPI only available option. The GDP data is seasonally adjusted. The Unit Root test (ADF) identifies that GDP series is the integrated of order one, so we use the growth rate of GDP, to make the series stationary (Table 1).

2. CPI

The data for the quarterly CPI is obtained by using monthly data, where we use arithmetic averages to calculate the quarterly values. In our analysis we use the change in the CPI, which is equal to the inflation rate in give period. The data is seasonally adjusted. The Unit Root test (ADF) shows that inflation is I(0) process, so we use it directly, as the series is stationary on levels (Table 1).

3. Refinancing rate of the Central Bank of Uzbekistan

The refinancing rate of the Central Bank of Uzbekistan is an interbank interest rate, used as a policy tool to signal the monetary policy stance and set by the Board in an annual basis. The rate of refinancing has not been changing over time quite often, thus the seasonal adjustment is not performed. The quarterly values are consists of the same rate set by the Board, and in cases when the rate is changed within one quarter, the simple average values are used. The Unit Root (ADF) test shows that refinancing rate is I(0) process, indicating the stationarity of data (Table 1).

4. Broad money (M2)

In our study we use broad money (M2) as another policy-related variable. The quarterly values are obtained by calculating arithmetic averages of monthly data. The data is seasonally adjusted and presented in real terms. The Unit Root test (ADF) identifies that M2 series is the integrated of order one, so we use the first difference, in order to make the series stationary (Table 1).

In the study carried out by Dabla-Norris and Floerkemeier (2006) narrow money (M1) is used as a policy-related variable, instead of the broad money (M2), pointing out the fact that the fluctuations in M2 can be caused not only by monetary policy, but also other factors. In our case the use of M1 is constrained by the data availability. Moreover, the abovementioned authors also state that the M1 is less correlated with output and prices, making an M2 still a valid choice.

5. Domestic credit

The data for the domestic credit used to capture the bank lending channel of the monetary policy. The quarterly values are obtained by using simple averages for monthly data. The Unit Root test (ADF) shows that domestic credit is I(2) process, so the second differences are used to make the series stationary (Table 1).

6. Real exchange rate

Throughout the study the nominal exchange rate is expressed as units of Uzbek sums per unit of the U.S. dollar. The real exchange rate index is calculated by adjusting the nominal exchange rate for the differentiation between domestic and the U.S. inflation (CPI is used), following Mankiw (2007). So, the increase of the index means the real depreciation, whereas the decrease indicates the real appreciation of the domestic currency.

The data is obtained from the web-site of the Central Bank of Uzbekistan, and quarterly values are calculated using arithmetic averages for the weekly data. So, final quarterly data should be seen as the period averages. The Unit Root test (ADF) shows that real exchange rate index is I(1) process, so we use the first differences, as I(1) type series should be differentiated once to make them stationary (Table 1).

7. The Russian GDP

In the paper by Dabla-Norris and Floerkemeier (2006), which we mainly follow, the authors use oil price index as a proxy for the development of remittances, which are one of main sources of foreign exchange. In our study we decided to modify this variable and use the Russian GDP instead of oil price index, as the GDP reflects the state of the economy better than the oil price or index. [7] Moreover, choosing the Russian GDP seems very reasonable if we take into account that remittance inflows from Russia account for 78 percent of the total remittances [8] .

The data is obtained f