Measuring monetary policy in a small open economy with managed exchange rates: the case of Taiwan.

• Author: Ho, Tai-kuang

1. Introduction

   In this article we use the sign restrictions method and specify a set of sign restrictions to identify monetary policy shocks in a small open economy. Most existing studies have adopted the recursive Cholesky approach to identify monetary policy shocks. This identification scheme may be appropriate for a large and relatively closed economy such as the United States. It is, however, much less likely to be valid for a small open economy (Cushman and Zha 1997). For a closed economy, the monetary transmission mechanism is viewed as operating through the interest rate. In the short run, a contraction in monetary policy is expected to increase the interest rate and thus depress private consumption and investment. This justifies the conventional practice to identify monetary policy shocks as innovations in the short-term interest rates. However, things are different for a small open economy. For a small open economy whose monetary authorities intervene heavily in the foreign exchange market, the effects of monetary policy operate through both the interest rate and the stock of foreign reserves. A contraction in monetary policy can take the form of either an increase in the interest rate or a decumulation of foreign reserves. This is the starting point of our identification of monetary policy shocks for a small open economy. We employ the sign restrictions methodology proposed by Uhlig (2005) to implement our identification scheme. The sign restrictions framework of Uhlig (2005) provides a conceptually clear way to impose identification conditions and to facilitate the comparison of our identification scheme to the other alternatives.

   Cushman and Zha (1997) and Berument (2007) also consider the identification of monetary policy shocks for a small open economy, taking Canada and Turkey, respectively, for their empirical studies. Cushman and Zha (1997) employ a structural vector autoregression (VAR) approach and include a broad set of foreign variables along with the domestic variables. Those authors' structural VAR relies on three conditions to identify monetary policy shocks. First, foreign variables are block exogenous to domestic variables. Second, domestic monetary policy reacts contemporaneously to the interest rate, monetary aggregate, and exchange rate, but not to real gross domestic product (GDP) and price level. Third, the exchange rate can respond contemporaneously to all domestic and foreign variables. These conditions allow simultaneous interaction among the monetary policy and a variety of domestic and foreign variables, which is absent in the recursive Cholesky approach. A common feature of our identification scheme and that of Cushman and Zha (1997) is that both of us allow simultaneous interaction between the interest rate and exchange rate (foreign reserves). The differences are, first, that we do not include foreign variables in our VAR model, so that the exogeneity of foreign variables with respect to domestic variables will not be an identifying condition in our case. Furthermore, we consider a small open economy with heavily managed exchange rates; whereas, that which Cushman and Zha (1997) consider is a small open economy with flexible exchange rates. Berument (2007) sticks to the recursive Cholesky approach, but he specifies a new variable as the policy instrument: the spread of the short-term interest rate and the depreciation rate of the exchange rate. Conventional method specifies a short-term interest rate as the policy instrument. Like Cushman and Zha (1997), Berument (2007) also considers the case of flexible exchange rates. Under flexible exchange rates, the effects of domestic monetary policy shocks on a small open economy revolve around the interest rate and the exchange rate effects. In the short run, a contractionary monetary shock is expected to increase the nominal interest rate and appreciate the value of domestic currency. That is why Berument (2007) uses exchange rate depreciation rather than foreign reserves as a part of the policy stance.

   Taiwan is used as a case study because Taiwan is a small open economy, and its monetary authorities accord a high weight to the stabilization of the exchange rate (Ho 2008). Existing studies tend to conclude that monetary policy has negligible effects on the Taiwanese economy. An extensive study by Lin (2003), for instance, finds that interest rate innovations identified from a VAR model have negligible effects on output and price level. The inability of the existing literature to find a strong effect of monetary policy provides us with a good chance to test our proposition: whether monetary shocks were really ineffective, or simply because they had not been correctly identified. We reexamine the effects of monetary policy shocks on real GDP and price level, using the above framework. Our principal findings are that a contractionary monetary policy shock causes a permanent and significant decline in real GDP, broad money, and price level. At the long-run horizon, monetary policy shocks account for about 20% of the variation of real GDP, broad money, and price level. Our identification scheme is able to avoid the puzzling impulse responses from which other identification schemes more or less suffer. Overall, these results support that our identification scheme is reasonable for a small open economy.

   The rest of the article is structured as follows. Section 2 describes the model specification and the sign restrictions to identify monetary policy shocks. Section 3 contains the main empirical results. We recover the monetary policy shocks identified from our method and report the impulse responses of the macro-variables to the monetary shocks. We provide a comparison of our method with the alternative identification schemes as well as the robust tests. The final section concludes. The appendix provides technical details of the sign restrictions methodology.

2. Model Specification and the Sign Restrictions

   We estimate a VAR model with six endogenous variables. The six variables are real gross domestic product (GDP), the short-term interest rate (IR), broad money (M2), the exchange rate (EX), the consumer price index (CPI), and foreign reserves (FR). (1) The choice of variables follows Lin (2003). Except for the interest rate where we use the level, all variables are transformed by logarithm. The data are at a quarterly frequency from the first quarter of 1981 to the last quarter of 2005. The VAR has four lags, no constant, and no time trends. (2) Real GDP, broad money, and CPI are seasonally adjusted. Christiano and Eichenbaum (1992) and Uhlig (2005) use non-borrowed reserves as the monetary aggregate. Here we employ M2 as the monetary aggregate because M2 growth is closely monitored by the monetary authorities of Taiwan, who have never targeted non-borrowed reserves as a strategy of money policy. Our data are obtained either from the Directorate-General of Budget, Accounting and Statistics, or from the Central Bank of Taiwan's website at http://www.cbc.gov.tw/.

   Because the focus of this article is the right sign restrictions to identify the monetary policy shocks, rather than the sign restrictions methodology itself, we leave the technical details of the methodology to the appendix. (3) We identify three shocks: a contractionary monetary policy shock, a positive aggregate supply shock, and a negative non-monetary aggregate demand shock. Table 1 shows the sign restrictions we employ to identify these shocks. The sign restrictions are imposed on each shock's impulse responses in the first K quarters following the shock. We have set K = 4, and therefore the restrictions are for horizons k, k = 0, 1, ..., 4. The conventional method to identify a contractionary monetary policy shock assumes that a contractionary monetary policy shock is associated with increases in the short-term interest rates or decreases in the monetary aggregate. Nevertheless, we argue that such an identification scheme, useful as it may be for a closed economy, is not suitable for identifying monetary policy shocks of a small open economy. Drawing from the wrongly identified monetary shocks, a researcher may conclude that monetary policy has no clear effect on real GDP. We argue that a small open economy, in which the monetary authorities are actively involved in the setting of short-term interest rates and exchange rates, may cut the liquidity provided to the market by raising the short-term interest rates or by selling off the foreign reserves. In terms of sign restrictions, we assume that a contractionary monetary policy shock does not lead to decreases in short-term interest rates and increases in the stock of foreign reserves for a certain period following a shock. Identifying monetary shocks in such a way is the main difference between the current article and the existing studies. Our identifications of aggregate supply and aggregate demand shocks are quite standard. A positive aggregate supply shock is defined as a shock that increases output and decreases the price level for four quarters following the shock. Negative non-monetary aggregate demand shocks are identified as shocks that decrease output, interest rate, and price level for four quarters following the shock. Standard theory suggests that a negative non-monetary aggregate demand shock is associated with a decrease in interest rates, and this offers a clear distinction to the contractionary monetary policy shocks.

   We are interested in the effects of monetary policy on real GDP and price level. We have imposed a minimum set of sign restrictions that are necessary to identify monetary policy shocks. No restrictions are imposed on the response of real GDP and price level to the monetary policy shocks. Thus, the question of interest is left agnostically open by the design of the...