The SVAR methodology

The VAR model assumes that Bangladesh’s economy can be represented by the following

structural equation:

A(L)Yt+α(L)Xt=εt, (2.1)

where,Ytis an n×1vector of endogenous variables andXtis a k×1vector of exogenous

foreign variables. A(L)is an(n×n)andα(L)is a(k×k)matrix polynomial lag operator.

εtis an(n×1)vector of structural disturbances with zero mean. It is assumed that shocks

are mutually uncorrelated. The reduced form of the structural model in Eq.(2.1) can be

written withp lags as

Yt=A1Yt−1+A2Yt−2+· · · +ApYt−p+αXt+et, (2.2)

where, α is a (n×k) matrix and et is a (n×1) VAR residuals. In the first stage, the

reduced form VAR(p) in Eq.(2.2) is estimated; next, we estimate the parameters in the

structural equation in several ways. The reduced form error and the structural shocks are

et=Bεt, (2.3)

where B is a non-singular (n×n) matrix that relates the VAR residuals (et) with the

structural shocks(εt). Multiplying both sides ofEq.(2.2) byB−1 yields

B−1Yt=D1Yt−1+D2Yt−2+· · ·+DpYt−p+B−1αXt+εt. (2.4)

Here,Dl=B−1Al for alll= 1, . . . , p.

Adding (In−B−1)Yt to both sides of Eq.(2.4)yields

Yt= (In−B−1)Yt+D1Yt−1+D2Yt−2+· · ·+DpYt−p+B−1αXt+εt, (2.5)

where,I is an(n×n)identity matrix. In a recursive VAR, theB matrix is lower triangular

and its diagonal elements are one.

The parametric restriction approach is applied in the SVAR model. FollowingDungey

& Pagan (2000a), the restrictions placed upon the system are of two types: First, we

assume the system as a whole is block recursive. As a small open economy, Bangladesh is

affected by various foreign variables such as international prices, but Bangladesh cannot

affect international prices. So, there are two blocks in the model: the first block contains

foreign variables that are important for Bangladesh, and the second block contains domestic

variables. The foreign block is placed ahead of the domestic block to ensure that the latter

does not enter the equations of the first block. Second, the recursive structure is assumed

inside each of the blocks. Finally, to ensure block exogeneity completely, following Zha

(1999) andDungey & Pagan(2000a), I restrict the domestic variables to affect the foreign

variables dynamically (in lag). I do not impose any other restriction in the lagged matrix,

Foreign block

Next, I discuss the relevance of the model variables separated into two blocks. The purpose

of using a foreign block is to explain movements in domestic variables and not the vice

versa. To be parsimonious with the available length of domestic time series, only two foreign

variables are included in the model. So the foreign block is represented asYf = (opw, pw)0

, where opwis the international oil price andpw is the CPI of Bangladesh’s major import

partner countries. Oil price is a commonly used variable in the monetary policy literature

and is considered a proxy for negative and inflationary supply shock (Kim & Roubini

2000), which also contains important business cycle information. Since Bangladesh is a

small economy, the CPI of other countries from where Bangladesh imports is assumed to

be important in determining domestic price movements. For example, the inflation rate in

India, one of Bangladesh’s largest trade partners, has correlations with the inflation rate

in Bangladesh (Bangladesh Bank 2014). In this study, the aggregate CPI for Bangladesh’s

nine major import partners has been considered based on their weights in total imports

over the entire period of study (2003-2014) and included as the foreign price variable. The

calculations of weights for each of the import partner countries are given in the Appendix

Table A4. The international oil price(opw)represents the supply side, and the foreign CPI

(pw)contains information on prices that Bangladesh’s exporters receive from Bangladesh’s

major trading partners. Hence, pw is an important component of foreign demand.

Domestic block

Given, the short period of study, deciding which domestic variables to include is a balance

between degrees of freedom and correct model specification. The commonly used variables

such as nominal interest rate (i), M2 money(m), nominal effective exchange rate (neer),

private sector is considered as the bank lending variable(cr)and the Consumer Price Index

(CPI) as the aggregate price(p)variable. Other important candidates, such as import price

and export price indices, are not included in the model due to their unavailability in the

required frequency for the entire period of study. The model also does not include any

variable relating to the asset price channel, such as the share price index or house price

index. The stock market index appear less important for Bangladesh, as can be seen from

statistics inTable 2.1and the economy-wide house price index is not available for the entire

period in the required frequency. These facts, along with the shorter time span, lead the

inclusion of six domestic variables mentioned above in the model.

The main objectives of the Bangladesh Bank as a monetary authority are to maintain

the stability of price, exchange rate and the overall financial system of Bangladesh, - this

provides the room for frequent intervention in the foreign exchange market. The argument

put forward by the ERPT literature on emerging economies is that macroeconomic vari-

ables have little explanatory power for exchange rates in the medium to short run (Zorzi

et al. 2007). For the baseline model I follow the structure provided by Zorzi et al. (2007)

and Bhattacharya et al.(2011) to order similar variables in the domestic block.

The order of the variables in the domestic block is as follows: Yd= (neer, i, m, cr, y, p)0. This ordering implies that the exchange rate captures the international effect first and

translates to all other variables including monetary policy rate. I place neer ahead of i,

assuming that the monetary policy decision is contemporaneously affected by exchange

rate but the policy rate cannot contemporaneously affect the exchange rate. The model

also assumes that domestic price is affected by external as well as all domestic variables.

Bhattacharya et al. (2011) use similar reasoning in ordering their SVECM model on the

Indian economy. This ordering also implies that financial sector variables affect real sector

variables contemporaneously, but not the other way around. The next subsection contains