The goal of modern central banks is to provide price stability. The phrase price stability is often associated with lower levels of price inflation. However, Relative Price Variability (RPV) also affects resource allocations and investment decisions and is associated with welfare losses. This paper assesses how the selection of monetary policy tool(s) affects this relative price variability.
Balke and Wynne (2007) argued that monetary policy shocks
increase the dispersion of a cross-section of relative prices, due to the non-neutrality
of money in the short term that may arise from different sources and that higher
dispersion of the cross-section disturbances creates adverse-supply side effects.
Central banks have various tools at their disposal to pursue their monetary
policies with. This study argues that the dispersion of the cross-section prices
will be different for different monetary policy tools and that this dispersion
is lower if a monetary authority chooses to mix policies in a Poole
(1970) sense, rather than choosing pure policies, such as interest rate
only or exchange rate only.
The most widely used tool of central banks is the short-term interest rate
(often the overnight interbank interest rate). Short-term rates are crucial
to economic performance (Bernanke and Blinder, 1992).
An increase in interest rates decreases investment expenditures and the demand
for consumer durable goods (Jones, 1954; Kau
and Keenan, 1980; Meltzer, 1974). Moreover, the durable-goods
sector is more interest sensitive than the non-durable goods sector (Erceg
et al., 2000). Furthermore, contractual monetary policies affect
crude goods prices more than processed goods prices (Balke
and Wynne, 2007). Thus, prices of different products are affected differently
when the short-term interest rate increases. Consequently, the price dispersion
among different types of products will be higher.
The second most-common tool that central banks of small open economies have
is the exchange rate, whereas large open economies may not utilize this tool
so effectively because the role of the exchange rate is small for the latter
group of economies. The presence of the exchange rate pass-through suggests
that higher levels of exchange rate (low values of domestic currency compared
to foreign currency) increase domestic prices. Kamin and
Rogers (2000) and Berument and Pasaogullari (2003)
claimed that after an increase in exchange rate, prices increase. The exchange-rate
sensitivity of tradable goods is greater than for non-tradable goods, which
causes a higher-level dispersion in relative prices after an increase in exchange
rate. Dunn (1970), Magee (1974) and
Takagi and Yoshida (2001) also claimed that the response of tradable goods
prices to a change in the exchange rate is quicker. In addition, the technology
shocks in the non-tradable sector and the demand shocks in the tradable sector
are strongly connected to the real exchange rate. Frenkel
and Taylor (2006) also noted that exchange rate influences price dispersion
between tradable and non-tradable goods, capital-intensive goods and labor-intensive
goods, as well as changing the costs of intermediate inputs and capital-intensive
goods. Thus, the conduct of exchange-rate-based policies affects prices of different
An increase in relative price dispersion leads to a reduction in the information
content of relative prices, which is often associated with the implementation
of more costly institutional arrangements such as a shorting of optimal contract
length and a consequent misallocation of resources, the result of which decreases
output (Kaul and Seyhun, 1990; Blejer
and Leiderman, 1980).
In order to analyze the effects of monetary policy tool selection on relative
price dispersion, we gathered data from Turkey. Turkey is an important emerging
economy and is the 17th largest economy in the world as of 2006. Turkey also
has conducted its monetary policies by using exchange rate, interest rate and
mixed policies for the time period that we consider (1998:2-2008:2). Thus, analyzing
the effects of various monetary policy tools on relative price dispersion for
Turkey is feasible. Turkey has well-developed and liberal markets with no price
controls or freezes during our sample period; thus, a change in relative prices
is due to market forces rather than administrative arrangements. Lastly, Turkey
has had high, persistent and volatile inflation rates without running into hyperinflation
for more than three decades. For present sample period, the average monthly
Consumer Price Index (CPI) inflation was 3.46 and the variance of the monthly
inflation was 7.45. This high volatility in inflation series decreases the chance
of a type II-error: not rejecting the null when it is false. A type II-error
is made when one does not reject the null hypothesis when, in fact, the null
is false. One method to decrease the incidence of type II-errors is to increase
the level of significance (α) or to decrease the confidence co-efficient
(1-α). However, Neter et al. (1985) also suggest
increasing the dispersion of the collected data. They show that increasing the
spacing decreases the standard errors of the parameters of interest and increases
the t-statistics for a given estimated parameter, the sample size and variance
of the errors. Since, Turkey has a high level of inflation compared to the most
other economies, spacing is higher and type II-error is lower (Berument
et al., 2005). The above factors make Turkey a unique country for
assessing any regularity of the effect of monetary policy tool selection on
the relative price dispersion.
CONDUCT OF TURKISH MONETARY POLICY
The second half of the 1980s was the era of economic liberalization for Turkey.
The interbank money market was created in 1986 and the Central Bank of the Republic
of Turkey (CBRT) started to engage in open-market operations in 1987. From the
beginning, the CBRT conducted its operations by using the exchange rate and
interbank interest rates simultaneously. During this period, the CBRT announced
the exchange rate every morning prior to the opening of the financial markets
and depreciated the basket daily by a fixed rate each month. Thus, economic
agents learned what the daily depreciation of the Turkish Lira (TL) would be
for the whole month after a couple of business days had passed; however, the
short-term interest rate fluctuated daily. Thus, the spread between the interbank
rate and depreciation assesses the excess return that the CBRT offers on the
Turkish lira above the depreciation rate-that is, the depreciation rate of the
TL against the US Dollar (USD). Therefore, spread can be evaluated as the opportunity
cost of holding TL against foreign exchange. However, this does not mean that
the CBRT controlled both instruments simultaneously. The monetary authority
may choose to control one of them and monitor the other (Berument,
2007). Turkish monetary authorities had been using a mixed policy, set up
by using both exchange rate and interest rates, to conduct their monetary activities.
At the beginning of 2000, Turkey adopted a new disinflation program. As part of this program, at the end of 1999, the CBRT announced daily (crawling) exchange rates for the 12 months of 2000. The CBRT also declared that the crawling peg would be abandoned in July 2001 and after this date, the crawling peg would be continued with an expanding band. We can claim that this to be an exchange-rate-based policy. Following the excess demand for foreign currency and the financial crisis that was triggered by a speculative attack on foreign currency on February 19, 2001, the CBRT let the exchange rate float. After March 2001, the CBRT started to use interest rates as its policy tool. After 2006, the CBRT adopted (explicit) inflation targeting and continued to use the overnight interest rate as a policy tool. Therefore, Turkey has utilized both types of pure policies as well as mixed policies to conduct its monetary activities.
DATA AND MODEL SPECIFICATION
The data is gathered from the Turkish Statistical Institute (TURKSTAT) and
the Central Bank of the Republic of Turkey, covering the monthly periods from
February 1988 to February 2008. Inflation is the logarithmic first difference
of the CPI as defined below:
where, πt is inflation and Pt is the consumer price
index at time t.
Inflation can be calculated by taking the logarithmic first difference or its growth from the earlier period as:
(1994) show that discrete changes of an index number is approximately equal
to the logarithmic first difference. Calculating inflation by taking the logarithmic
first difference is quite common in study; for example, Holden
and Driscoll (2003), Banerjee et al. (2007),
Siklos and Abel (2002) and Reitz and Taylor (2008).
The CPI has seven basic components:
||Group index of clothing
||Group index of culture
||Training and entertainment
||Group index of food-stuffs
||Group index of home appliances and furniture
||Group index of medical health and personal care
||Group index of housing and group index of transportation and communication
However, in 2005, TURKSTAT changed the calculating method of the CPI, discontinuing
some of the subcomponents of the seven CPI subgroups; hence we could gather
from the main seven groups only. Inflation for each subgroup is defined as:
where, πi, t is the inflation of each subgroup and Pi,
t is the price index for each subgroup.
Fischer (1981), Smith and MacKinnon
(1987) and Chang and Cheng (2002) calculated relative
price variability (V) using Eq. 1 and 2
where, wi is the ith weight of the group index in CPI.
Berument (2007) spread had been taken as the difference
between the monthly simple interbank rate and monthly exchange as a measure
of monetary policy. In particular, spread is the difference between the monthly
interest rate return and the depreciation of the TL against the USD. The growth
rate of industrial production is y and exchange rate (depreciation) is the logarithmic
first difference of the TL value of the USD. The exchange rate, interbank rate
and industrial production are gathered from the CBRT electronic data delivery
system; price indices are gathered from TURKSTAT.
Before going over the estimation specification, we plot relative price variability (V) and seasonally adjusted real Gross Domestic Product (GDP) growth series to see if there is any pattern between these two series, shown in Fig. 1. Figure 1 suggests that especially after 1993, this relationship is negative. The correlation coefficient for the full period is -0.10. This correlation coefficient is -0.20 after 1992 and after April 1994 and February 2001, when economic crises and other turbulences were a crippling shock to the system, GDP growth diminishes, but the relative price variability increases.
In order to explore the effect of monetary policy shocks on relative prices
we set up a five-variable VAR model. The VAR specification includes y, exchange
rate, spread, V and Inflation (benchmark VAR specification). It is important
to recognize that the exchange rate enters the VAR specification twice: once
as an exchange rate and once as the difference between the interbank rate and
the percentage change in the exchange rate. This might be considered a problem.
Here, we impose the constraint that the difference between the interbank rate
and the depreciation rate can be used as an indicator of monetary policy and
we treat the interbank rate above the depreciation rate as a variable separate
from the exchange rate. Entering different interest rate spreads along with
their components is also common in the literature (Bernanke,
1990; Friedman and Kuttner, 1992) or the difference
of a series along with its level can be used (Bernanke, 1983).
When the VAR analysis were performed, each equation had 11 monthly dummies
to account for seasonal changes as well as dummies for the 1994 and 2001 financial
crises, for the time span from 1988:02 to 2008:02.
||Quarterly graphs of relative price volatility and GDP growth
The lag length of the VAR specification is determined by Akaike Information
Criteria (AIC) and Bayesian Information Criteria (BIC), which suggest that the
lag length should be 1. Once the model(s) is estimated, the impulse responses
are gathered with the Generalized Impulse Responses (GIR), as suggested by Pesaran
and Shin (1998). The GIR produces impulses that are insensitive to ordering
of variables. Similar to Berument (2007), we also calculate
the impulse responses with the Chelosky decompostion and the results are similar.
In order to explore the effect of mixed policies that are measured by spread
innovations compared to exchange rate (depreciation) and interbank rate, we
utilized two sets of VAR specifications. Because spread is the linear combination
of exchange rate and interbank rate, we could not include these three variables
simultaneously, but only two of them at a time. Figure 2a
and b show the responses of volatility (V) to one-standard
deviation shock in the exchange rate and spread and Fig. 3a
and b show the responses to one-standard deviation shock
in the interbank rate and spread. The Bayesian simulation method, with replications
of 500 iterations, is used to obtain the confidence intervals for the impulse
response functions. In our model, the impulse response functions are gathered
from Monte Carlo simulations, in which the confidence intervals of impulse responses
are ±2SE bands. In Fig. 2, 3, the
middle lines show the impulse response functions and the upper and lower lines
represent the confidence intervals. When the confidence interval contains the
horizontal line, the null hypothesis that there is no effect of any of the shocks
that we consider on RPV cannot be rejected. Hence, adding the horizontal line
for that particular period shows evidence of statistical insignificance.
Shock to exchange rate increases V contem- poraneously. It reaches its peak
at the second period and higher V is observed for the 10 periods that we consider;
this effect is statistically significant. When we introduce a one-standard deviation
shock to spread, V decreases immediately. The effect reaches its bottom in the
second period but lower V persists for the 10 periods that we consider. Moreover,
this effect is statistically significant. In order to assess how the interbank
rate and spread contribute to RPV, we estimate another five-variable VAR model
by replacing exchange rate (depreciation) with the interbank interest rate (the
alternative VAR specification). Figure 3 shows how interbank
rate and spread affect V. Similar to the impulse responses shown in Fig.
2, innovation in spread decreases V. A shock to the interbank rate increases
V and it reaches its peak in the second period; again this effect persists for
||Impulse response functions, (a) response of V to exchange
and (b) response of V to spread
||Impulse response functions, (a) response of V to spread and
(b) response of V to interbank
||Impulse responses for the benchmark specification. Response
to generalized one SD innovations ±2SE
||Impulse responses for the alternative specification. Response
to generalized one SD innovations ±2SE
||Forecast error variance decompositions for the Benchmark specification
|*, **Indicate significance at the 0.10 and 0.05% level, respectively
||Forecast Error Variance Decompositions for the Alternative
|SE are reported in columns under the corresponding coefficient.
*,** Indicate significance at the 0.10 and 0.05% level, respectively
Figure 4 and 5 also report the
impulse response functions when a one-standard deviation shock is given to each
variable and how the other variables react to it. Pure policies increase but mixed policies decrease the relative price variability.
The dynamic effects of monetary policy shocks are gathered by impulse response
functions obtained from two VAR estimates. However, in order to investigate
the contribution of monetary policy shocks to the volatility of various economic
aggregates, the forecast error variance decompositions have been estimated.
These help to assess whether monetary policy shocks have been an important independent
source of shocks for RPV, inflation and output. Table 1 and
2 show the forecast error variance decompositions for these
two five-variable VAR specifications. 6, 12, 18, 24 and 30 step-ahead forecast
error variance decompositions are reported for the variables in interest.
Table 1 suggests that RPV explains approximately 1.2% of the variability of output and 11% of the variability of inflation. Even if these explanatory powers for output and inflation may be considered low, RPV has more power to explain the behavior of output than the inflation rate and spread for output and inflation. Similarly, PRV explains the behaviour of inflation more than exchange rate and spread. Considering the importance attributed to exchange rate and interest rate that affects spread in the literature, the role of relative price volatility is something that we cannot ignore. Table 2 shows the estimates of the same analyses but uses interbank rate rather than spread and basically reveals the same information: The contribution of RPV cannot be dismissed in output and inflation.
In addition, exchange rate volatility has more power to explain the volatility
of output and inflation than the interbank rate does. This makes sense, as Turkey
is a small open economy with a small financial market, compared to the European
Union average (financial deepness = Total asset of Turkish financial assets/GDP
was 116% as of September 2007 and 270% for the European Union). RPV is mostly
explained with lag values of inflation volatility and this parallels Parks
(1978), Fischer (1981) and Chang
and Cheng (2002).
Central banks have various tools in their arsenals to conduct their monetary policies. Each of these tools affects prices of different products differently. Using only one tool may require the prices of a set of products to adjust too much, compared to prices of other products whose prices are less sensitive. On the other hand, using mixed policies involving more than one policy tool may affect the prices of a bigger pool of products; the required adjustments will be lower on the prices of those products and thus may not increase relative prices as much. This study provides empirical evidence that interest-rate-only and exchange-rate-only policies increase the relative price variability and using interest and exchange rates simultaneously decreases the relative price variability; thus, to stabilize the economy, mixed policies might be preferred to pure policies.