Currency Intervention A Case Study of An Emerging Market

Journal of International Money and Finance 37 (2013) 2547
Contents lists available at SciVerse ScienceDirect
Journal of International Money and Finance

journal homepage: www.elsevier.com/locate/jimf
Currency intervention: A case study of an emerging marketq

Rene A. Fry-McKibbin a, *, Sumila Wanaguru a, b
a b
Centre for Applied Macroeconomic Analysis (CAMA), Crawford School of Public Policy, The Australian National University, Canberra, ACT 0200, Australia Central Bank of Sri Lanka, Sri Lanka
a b s t r a c t
JEL classication: F31 F36 F41 Keywords: Foreign exchange intervention Currency intervention Exchange rate volatility Reserve accumulation Factor model Emerging markets
Using a unique dataset on daily foreign exchange intervention and a new methodological framework of a latent factor model of central bank intervention, this paper addresses the effects of intervention in an emerging market. Events in nancial markets from 2002 to 2010 provide a natural experiment to evaluate the short and medium term objectives of the central bank to contain excessive exchange rate volatility and to accumulate foreign reserves respectively. In the low volatility period in the rst part of the sample, the central bank is successful in inuencing the currency when pressure is to appreciate, accumulating international reserves. The same model estimated for the global volatility period in the second part of the sample shows the central bank intervening to mitigate excessive exchange rate volatility in line with the short-term objective. The results point to the need to consider the cross currency market interdependence between emerging markets when modeling intervention. 2013 Elsevier Ltd. All rights reserved.
1. Introduction The motives for central banks to intervene in the foreign exchange market include reducing the economic costs associated with exchange rate volatility which affects international trade, nancial
q We are grateful for the useful comments provided by the referee of the paper. Wanaguru acknowledges support from the Central Bank of Sri Lanka. The views expressed in this paper are those of the authors and do not necessarily represent the views of the Central Bank of Sri Lanka. McKibbin acknowledges support from ARC Discovery Project DP0985783.
* Corresponding author. Tel.: 61 2 6125 3387. E-mail address: renee.mckibbin@anu.edu.au (R.A. Fry-McKibbin). 0261-5606/$ see front matter 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jimonn.2013.05.007
26
R.A. Fry-McKibbin, S. Wanaguru / Journal of International Money and Finance 37 (2013) 2547
ows, foreign investment and economic growth, and accumulating international reserves to strengthen a countrys macroeconomic fundamentals (Szakmary and Mathur, 1997; Sarno and Taylor, 2001; Disyatat and Galati, 2007; Pointines and Rajan, 2011). These objectives are particularly important for emerging markets as they are more prone to and affected by external shocks than their developed counterparts. Meanwhile, accumulating international reserves helps to establish the condence of foreign investors in the domestic economy by positively affecting sovereign risk, and vulnerability to external shocks can be alleviated through a high level of reserve adequacy (Mulder and Perrelli, 2001; Dominguez et al., 2011). Using a unique dataset on daily foreign exchange intervention and a new methodological framework, this paper addresses the effects of intervention on exchange rate volatility and reserve accumulation for emerging markets using Sri Lanka as an example.1 The ofcially announced intentions of the Central Bank of Sri Lanka are exactly those mentioned above but with a time frame associated with each objective in that in the short term, intervention is to contain excessive volatility in the exchange rate, and in the medium term is to accumulate international reserves (Central Bank of Sri Lanka, 2007).2 Determining the effects of intervention for emerging markets is constrained by data availability and with the exception of Disyatat and Galati (2007) for the Czech Koruna, there are few published works in this area.3 Sarno and Taylor (2001) and Disyatat and Galati (2007) provide good surveys to evaluate intervention and its effects on exchange rate volatility with the conclusions tending to be that intervention can be effective and is conducted mainly in response to a rapidly appreciating domestic currency. From a reserve accumulation perspective, the large stocks of reserves held by emerging markets is now attracting attention following the economic and nancial market collapses of the last ve years. Important papers examining this issue include Dominguez et al. (2011) and Dominguez (2010). A framework which naturally lends itself to modeling central bank foreign exchange intervention but which has not previously been applied to this topic is the latent factor framework.4 This class of models is often used to calculate volatility decompositions to decompose nancial market asset returns into specied sources of volatility associated with the factor structure such as global, domestic, asset market or country factors (Diebold and Nerlove, 1989; Mahieu and Schotman, 1994; Dungey, 1999). This paper constructs a factor model of intervention for a set of daily currency returns of Sri Lanka and its major trading partners as well as Sri Lankan intervention data which is modeled endogenously. The weight placed on the objectives of a central banks intervention policy at any point in time is a function of the prevailing external global economic environment, the domestic economic environment including policy regime choices, as well as the general level of development of a country. Our model reects this environment for an emerging country by specifying each Sri Lankan and trading partner currency return as a function of global, domestic and intervention factors. The global factor affects all currency returns in the model but allows each market to respond in different ways. It captures movements external to the domestic economy and encompasses concepts such as but not exclusively global market fundamentals, global liquidity conditions and general trader risk aversion. A domestic factor is specied for each variable and captures movements specic to each market. Intervention is also a function of global and domestic factors. Using the fact that it is known on which days intervention policy is enacted, an additional intervention factor is specied for the Sri Lankan currency equation which shares features of the net intervention equation. This relationship exists only on days on which the central bank intervenes and the feature of known intervention days is also used as part of the identication of the model. Events in nancial markets in the sample period from January 2002 to December 2010 provide a natural experiment to evaluate the short and medium term objectives of the Central Bank. The model is
We are grateful to the Central Bank of Sri Lanka for providing us with all data, particularly the intervention data. Intervention in Sri Lanka is not aimed at targeting an exchange rate level (Central Bank of Sri Lanka, 2007), implying that the intervention strategy is to lean against the wind to reduce exchange rate volatility. 3 Emerging country central banks and organisations such as the International Monetary Fund aim to ll this gap (Pattanaik and Sahoo, 2001; Mandeng, 2003; Guimares and Karacadag, 2004; Herrera and Ozbay, 2005; Kamil, 2008; Adler and Tovar, 2011). 4 The manuscript, Aruman (2003) considers intervention in a latent factor framework but uses a factor structure different to that adopted here.
2
27
estimated over two periods. The rst corresponds to the relative calm and low volatility of nancial markets in the rst half of the sample, and the second to the period of global volatility associated with the global nancial crisis in the second half. If the commitment to the medium term objective of reserve accumulation and the short term objective of reducing exchange rate volatility is met, the volatility decompositions for each period should differ. The data provided distinguishes between days of intervention through net sales and net purchases of US dollars providing further evidence on the commitment of the central bank to each objective. The results suggest that the central bank is successful in achieving its short-term and medium-term objectives of containing exchange rate volatility and accumulating reserves. In the low volatility period, the central bank tends to intervene in response to global rather than domestic factors and is able to inuence overall foreign exchange return volatility by 5.5 percent. Splitting the data into days of intervening through purchases versus sales of US dollars shows that intervention is most effective when the bank purchases US dollars. This suggests that the central bank is successful in inuencing the exchange rate when the pressure in currency markets is to appreciate the Sri Lankan rupee, hence accumulating international reserves in line with their medium-term target as is expected during a period of calm. The same model estimated for the global volatility period presents strikingly different results. Sales of US dollars is important during this time with the central bank intervening to mitigate the exchange rate volatility in line with the short-term objective. The rest of the paper proceeds as follows. Section 2 presents the exchange rate and intervention data used in the empirical application. The modeling framework is developed in Section 3, and Section 4 discusses the GMM methodology adopted to estimate the models of intervention. Section 5 presents the empirical results, rst focusing on the low volatility period and later the global volatility period. This section also considers the role of emerging market factors and explores some of the interdependencies between the regional currency markets on non-intervention compared to intervention days. Section 6 concludes. 2. Exchange rates and intervention This section presents a preliminary analysis of the data used in the model of foreign exchange intervention in Sri Lanka. The data comprise of n 5 daily exchange rate returns of the euro (EURt), the Indian rupee (INRt), the Japanese yen (JPYt), the British pound (GBPt) and the Sri Lankan rupee (SLRt), expressed in US dollars, as well as daily net foreign exchange purchases by the Central Bank of Sri Lanka (INTt). Exchange rate returns are computed by taking the rst difference of the natural logarithm of the exchange rates and multiplying them by 100. Net foreign exchange purchases are in millions of US dollars. All series are demeaned and scaled by their respective standard deviations, and are expressed in standardized units. The exchange rates and returns are shown in Fig. 1.5 An increase in the value of the exchange rate indicates an appreciation of the US dollar against the local currency. The selection of exchange rates corresponds to Sri Lankas main trading partner countries according to the weights assigned by the Central Bank of Sri Lanka in calculating the 24-currency real effective exchange rate. The top six countries trade weights in the calculation of the real effective exchange rate are the US (19.74 percent), India (15.57 percent), the UK (9.86 percent), China (6.41 percent), Germany (5.88 percent) and Japan (4.99 percent).6 The Chinese yuan is excluded from the model as the focus is on countries with oating exchange rate regimes. The sample consists of a selection of developedmarket exchange rates, as well as the emerging-market exchange rate of the Indian rupee. The Indian rupee provides a convenient point of comparison with the Sri Lankan rupee in the model. The intervention data is net foreign exchange purchases data for Sri Lanka as shown in Fig. 2 and is plotted against the log of the level of the Sri Lankan rupee in panel (a), and the percentage returns in (b). Positive values of the intervention series represent purchases of US dollars and negative values
5 The outliers in the euro on March 1 and 2, 2005, as depicted in Fig. 1, are removed by regressing the euro returns against a dummy variable for each outlier. 6 See Box Article 12 of the Central Bank of Sri Lanka Annual Report 2010, Revision of Effective Exchange Rate Indices, http:// www.cbsl.gov.lk/pics_n_docs/10_pub/_docs/efr/annual_report/AR2010/English/9_Chapter_05.pdf (accessed June 5, 2011).
28
Euro .2 .1 .0 -.1 -.2 -.3 -.4 2004 2009 2003 -.5 4.0 0.0 -4.0 -8.0 8.0
Euro returns
2005
2007
2002
2003
2005
2006
2007
2010
2002
2004
2006
2008
Indian rupee 4.0 3.9 3.9 3.8 3.8 3.8 3.7 3.7 3.6 8.0 4.0 0.0 -4.0 -8.0
Indian rupee returns
2005
2008
2002
2003
2005
2006
2007
2008
2009
2003
2007
2009 2009 2009
2009
2004
2010
2002
2004
2006
Japanese yen 5.0 4.9 4.8 4.7 4.6 4.5 4.4 2003 2006 2007 2010 4.3 4.0 0.0 -4.0 -8.0 8.0
Japanese yen returns
2005
2006
2007
2008
2002
2005
2008
2009
2003
2004
2002
2004
British pound -.3 -.4 -.5 0.0 -.6 -.7 2005 2010 2006 2002 2003 2004 2007 2008 2009 -.8 -4.0 -8.0 8.0 4.0
British pound returns
2002
2003
2004
2007
2005
2006
2008
2008
Fig. 1. Daily Log Exchange Rates and Percentage Exchange Rate Returns, January 2002December 2010. Notes: Returns are for the euro, the Indian rupee, the Japanese yen and the British pound against the US dollar. The shaded area indicates the period of global volatility from July 2, 2007December 31, 2010. (Source: Central Bank of Sri Lanka).
2010
2010
2010
2010
29
4.8 4.8 4.7 4.7 4.7 4.6 4.5 2002 2003 2005 2006 2007 2008 2009 2004 2010 4.5
120 80 40 0 -40 -80 -120 -160
1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 -2.0 2003 2004 2005 2006 2007 2002 2008 2009 2010 -2.5
120 80 40 0 -40 -80 -120 -160
(a)
SLR/USD exchange rate (log) (left) Net foreign exchange intervention USD mn. (right)
(b)
SLR/USD exchange rate returns (%) (left) Net foreign exchange intervention USD mn. (right)
Fig. 2. Sri Lankan Rupee Exchange Rate and Intervention Data, January 2002December 2010. Notes: Panel (a) is the daily log exchange rate against the US dollar and net foreign exchange intervention (USD mn). Panel (b) is the percentage daily exchange rate returns against the US dollar and net foreign exchange intervention (USD mn). The scale on the left of each panel is for the exchange rate variables and on the right is the intervention data. The shaded areas indicate the period of global volatility from July 2, 2007 December 31, 2010. (Source: Central Bank of Sri Lanka).
represent sales. Net daily foreign exchange purchases are conducted only in the US dollar market; however, by doing so, the central bank indirectly inuences the exchange rates of other currencies against the Sri Lankan rupee. The sample extends from January 1, 2002 to December 31, 2010, and is chosen to begin based on the availability of the daily foreign exchange intervention data after the oating of the Sri Lankan rupee in 2001. For estimation of the model for the low volatility period in Sections 5.15.3, the sample period is chosen to end on June 29, 2007. The model for the global-volatility period in Section 5.4 is estimated from July 2, 2007 to December 31, 2010. The global volatility period is illustrated by the shaded area in the gures. The separation between the low and global volatility periods is chosen based on reading the minutes and press releases of the Board of Governors of the Federal Reserve System on the assumption that the trigger for the global nancial market volatility began with the emergence of the US subprime mortgage market distress. The press release dated June 28, 2007 which is the day prior to the end of the low volatility period shows that the Fed chose not to change interest rates and did not specically comment on the vulnerabilities to US growth or the nancial system at that time (Board of Governors of the Federal Reserve System, 2007c). The rst sign of concern by the Fed about the downside risks to the economy emerged just over a week later on August 7, followed by the provision of liquidity to nancial markets on August 10 (Board of Governors of the Federal Reserve System, 2007a,b). Hence, the beginning of the end of the low volatility period is chosen to be June 29, the day following the press release on June 28 where no difculties were formally agged to the public.7 Table 1 presents descriptive statistics on the rupee and other exchange rate returns in the model for the total sample period, as well as for low volatility and global volatility periods for all days, days of intervention and days on which there is no intervention. Table 2 presents similar statistics for the net intervention data.8 The central bank does not dene the meaning of excessive volatility when
7 Conditional structural break tests for a break of unknown timing are also performed for each currency return in a VAR of all currency returns based on the method of Hansen (2000). The results indicate the structural breaks for each currency return series are 17 October, 2008 for the euro, 1 September, 2008 for the Indian rupee, 7 March, 2008 for the yen, 17 October 2008 for the pound, and 12 April, 2007 for the Sri Lankan rupee. All currency returns apart from that for Sri Lanka show a structural break in 2008 either before or just after the collapse of Lehman brothers. The choice of high volatility period is chosen to be that corresponding to the subprime period prior to the extreme turmoil leading up to the Lehman brothers collapse as justied by the anecdotal economic events described. 8 Note that changes in cross rates for example between the Sri Lankan rupee and the euro, are not formally modeled in this paper, with all exchange rates expressed against the US dollar.
30
Table 1 Descriptive statistics of the exchange rate returns (percent). Notes: the exchange rates are expressed in terms of US dollars. The statistics are calculated for each sub-period for all days (All), days on which there is no intervention (Non-int) and days on which there is intervention (Int). The total sample period is from January 1, 2002December 31, 2010, the low volatility period is from January 1, 2002June 29, 2007, and the global volatility period is from July 2, 2007December 31, 2010. Statistics Variable Total sample All No. of obs. Max EUR INR JPY GBP SLR EUR INR JPY GBP SLR EUR INR JPY GBP SLR EUR INR JPY GBP SLR EUR INR JPY GBP SLR EUR INR JPY GBP SLR 2171 7.53 3.40 4.41 4.68 1.25 6.69 3.25 4.73 4.58 2.04 0.02 0.00 0.02 0.00 0.01 0.75 0.43 0.69 0.69 0.17 0.359 0.272 0.194 0.239 2.088 8.706 11.833 6.515 7.563 32.522 Non-int 1089 7.53 1.93 3.44 2.61 1.15 6.69 3.25 2.71 2.71 2.04 0.01 0.01 0.02 0.01 0.01 0.77 0.44 0.64 0.64 0.21 0.694 0.714 0.263 0.041 2.324 10.710 10.876 7.365 4.342 29.777 Int 1082 3.65 3.40 4.41 4.68 1.25 4.73 2.58 4.73 4.58 1.06 0.03 0.00 0.06 0.01 0.01 0.73 0.42 0.73 0.74 0.14 0.006 0.003 0.128 0.335 1.346 6.594 11.790 5.748 8.837 30.738 Low volatility period All 1324 7.53 1.93 3.44 2.61 1.15 6.69 2.26 2.71 2.46 2.04 0.03 0.01 0.01 0.02 0.01 0.69 0.31 0.59 0.56 0.19 0.589 0.538 0.093 0.028 2.332 11.668 12.700 6.415 4.481 27.569 Non-int 763 7.53 1.93 3.44 2.61 1.15 6.69 2.26 2.71 2.46 2.04 0.03 0.01 0.02 0.02 0.01 0.75 0.35 0.61 0.60 0.21 0.971 1.257 0.037 0.030 2.702 15.048 12.437 7.252 4.588 28.224 Int 561 2.55 0.98 1.29 1.47 0.77 2.19 1.57 2.56 2.08 1.06 0.04 0.02 0.04 0.03 0.01 0.59 0.24 0.55 0.50 0.15 0.117 0.627 0.394 0.014 1.032 4.393 13.086 4.047 4.004 16.972 Global volatility period All 847 3.65 3.40 4.41 4.68 1.25 4.73 3.25 4.73 4.58 1.67 0.00 0.01 0.05 0.03 0.00 0.84 0.57 0.82 0.86 0.16 0.202 0.219 0.201 0.266 1.404 6.163 8.213 5.704 6.870 45.890 Non-int 326 2.99 1.74 2.47 2.39 1.14 3.80 3.25 2.21 2.71 1.67 0.02 0.00 0.02 0.01 0.00 0.81 0.58 0.72 0.75 0.20 0.351 0.407 0.546 0.099 1.052 5.436 7.425 6.717 3.745 34.851 Int 521 3.65 3.40 4.41 4.68 1.25 4.73 2.56 4.73 4.58 1.04 0.01 0.02 0.09 0.04 0.00 0.85 0.56 0.88 0.93 0.12 0.351 0.407 0.546 0.099 1.052 5.436 7.425 6.717 3.745 34.851
Min
Mean
Std. dev.
Skewness
Kurtosis
dening its objectives in relation to intervention, hence, there is no formal rule governing when intervention should occur. However, Fig. 2 and Table 2 show that the Central Bank of Sri Lanka intervenes frequently. Over the sample period, intervention takes place on approximately 50 percent of all days, with a fairly even split between net purchases (547 days) and sales (535 days). There is more intervention in the global volatility period (62 percent of days) compared to the low volatility period (43 percent of days). Fig. 2 and Table 2 show that the magnitude of volatility for the intervention data is greater during the global volatility period. The standard deviation of intervention increased from 7.81 million in the low volatility period to 20.03 million in the global volatility period. During this time the Central Bank of Sri Lanka went from being a net seller to a net purchaser as shown by the mean across the sub-periods. For Sri Lanka, the increased volatility probably results from domestic and external reasons. This period not only coincides with the global nancial crisis, but also the nal phase of the 25 year civil war in Sri Lanka. Turning again to Table 1, on days during the global volatility period when the central bank intervenes, volatility is higher on intervention days than on days of no intervention for all countries excluding India and Sri Lanka. For example, on non-intervention days, the standard deviation of the euro returns is 0.81 percent compared to 0.85 percent on the intervention days, and increases from 0.75 percent to 0.93 percent for the pound. In contrast, the standard deviation for the Sri Lankan rupee returns falls from 0.20 percent to 0.12 percent perhaps suggesting that the central bank is effective in containing exchange rate return volatility through intervention when volatility is high, or perhaps reecting the improvement of domestic conditions.
31
Table 2 Descriptive statistics for the intervention data. Notes: the intervention data are expressed in millions of US dollars. The statistics are calculated for each sub-period on all days there are purchases or sales of US dollars (Intervention), days on which US dollars are purchased (Purchases) and days on which US dollars are sold (Sales). The total sample is from January 1, 2002December 31, 2010, the low volatility period is from January 1, 2002June 29, 2007, and the global volatility period is from July 2, 2007 December 31, 2010. Statistics Total sample No. of obs. Mean Max Min Std. dev. Low volatility period No. of obs. Mean Max Min Std. dev. Global volatility period No. of obs. Mean Max Min Std. dev. Intervention 1082 0.11 99.75 118.45 14.99 561 0.35 39.15 40.00 7.81 521 0.60 99.75 118.45 20.03 Purchases 547 9.71 99.75 0.25 11.60 307 4.80 39.15 0.25 4.89 240 16.00 99.75 0.50 14.36 Sales 535 9.68 0.25 118.45 11.27 254 6.60 0.25 40.00 5.89 281 12.47 0.25 118.45 13.91
The preliminary statistics particularly for skewness in the second panel of Table 1 are also pertinent to examine. Comparing the skewness statistics for the low and global volatility period for the dataset shows that the sign of skewness for most currencies are negative in both the low and the global volatility periods. The striking exception is for the pound sterling return, which during the low volatility period the skewness values are all negative (as are most of the currency returns in the model). During the global volatility period the sign of skewness for the pound is positive for all types of days. As shown Harvey and Siddique (2000), risk averse investors prefer positive skewness to negative skewness in asset returns. It is often the case that values of skewness switch signs to become positive during nancial market crises as lower returns are accepted in exchange for positive skewness (Fry et al., 2010). Only skewness in the sterling responds in this manner, indicating that the sterling is acting as a safe haven currency in the global volatility period. It should be noted that foreign exchange intervention in the broad sense includes measures which affect the exchange rate both directly and indirectly. Intervention from a more narrow perspective is restricted to purchases or sales of foreign exchange by a monetary authority with a view to inuencing the exchange rate directly. This paper analyzed the narrow perspective of intervention. No stance is taken on whether intervention here is sterilized or not, although it is probably partially sterilized, as Canales-Kriljenko (2003) points out, in contrast to developed countries it is not common for central banks in emerging countries to fully sterilize. 3. Model specication The analytical framework employed in this paper is a latent factor model of exchange rate returns in the tradition of Mahieu and Schotman (1994), Diebold and Nerlove (1989) and Dungey (1999), where exchange rate returns are presented as functions of a set of independent latent factors. The factors in this application capture movements that are common to all exchange rate returns (global factors), idiosyncratic to each asset (domestic factors), and related to intervention (intervention factors). Adopting a factor structure has several advantages. First, the approach provides a parsimonious representation of the data. Second, observable variables do not have to be identied or modeled. Third, the approach is convenient to use, as the model implicitly takes into account all disturbances affecting the system of exchange rate returns. Finally, iid and unit variance assumptions on the factor structure allow the decomposition of exchange rate returns into the
32
contribution that each of the factors makes to overall volatility. The volatility decompositions are the main vehicle for analysis. In nalizing the factor model of central bank intervention, the model is built up in two stages. Section 3.1 species a factor model of exchange rate returns without formally modeling the effect of intervention. However, the model distinguishes between non-intervention and intervention days. On non-intervention days, exchange rate returns are a function of global and domestic factors. On intervention days, the exchange rate returns are a function of the same factors; however, the effect of each factor on each exchange rate return, as given by the factor loadings, is allowed to change through the formal modeling of structural breaks. These are designed to capture changes in the global and domestic dependence structures among the exchange rate returns which may be prevalent on the days that a central bank chooses to intervene in the foreign exchange market. The modeling of only the exchange rate returns in the rst stage is to provide a sense of these dependence structures before the formal introduction of intervention. In the second stage of modeling described in Section 3.2, care is again taken to distinguish between non-intervention and intervention days. The intervention variable is introduced into the model of exchange rate returns and follows the same factor structure as the exchange rates, in that it is specied as a function of a global and an idiosyncratic factor. However, on intervention days the Sri Lankan rupee exchange rate returns are allowed to be a function of the idiosyncratic factor associated with the intervention data, effectively specifying an additional domestic factor (an intervention factor) for the two Sri Lankan equations on these days. This model is able to provide evidence on the effectiveness of intervention by the Central Bank of Sri Lanka, as the contribution of intervention to the volatility of the Sri Lankan exchange rate returns in comparison to the global and idiosyncratic factors is able to be assessed. 3.1. A factor model of exchange rate returns This section species a latent factor model of exchange rate returns for the intervention and nonintervention days, while suppressing the formal role for intervention. The model consists of n zero mean daily bilateral exchange rate returns expressed against the US dollar. The data-set is separated into two parts, which aids identication as outlined in the discussion of the Estimation Method in 1 Section 4. Let e0 t denote the exchange rate returns on non-intervention days (j 0), and let et denote the exchange rate returns on intervention days (j 1), such that
et
EURt ; INRt ; JPYt ; GBPt ; SLRt
j 0; 1:
(1)
3.1.1. Non-intervention days The dynamics of the ith exchange rate returns on non-intervention days (j 0) is governed by a set of independent latent factors
0 0 e0 i;t li wt gi ui;t i 1; 2; .; n;
j 0:
(2)
The global factor in the model (wt) captures common shocks affecting each of the n exchange rate 0 returns in the model with their own parameter loading li .9 The domestic factor ui,t captures shocks specic to each currency market, and reects own-country fundamentals that are independent of global conditions. The loadings on the idiosyncratic factors are g0 i: 3.1.2. Intervention days On intervention days (j 1), it is assumed that there is a possibility of higher volatility in the exchange rate market, perhaps prompting intervention. To allow for this, structural breaks in the factor
9 An alternative structure is to formally model a common numeraire factor to show that all returns are expressed in US dollars. This factor would affect each exchange rate return with a xed loading in each equation. The presence of the numeraire factor imposes a no-arbitrage condition on the model, as shown in Dungey (1999). However, computationally this specication did not work for this application.
33
structure are specied for intervention days. The dynamics of exchange rate returns for intervention days (j 1) is expressed as
e1 i;t
1
1 0 1 l0 i li wt gi gi ui;t i 1; 2; .; n;
j 1
(3)
where li and g1 i are the structural breaks in the parameters on the global and idiosyncratic factors. In matrix form, the model of exchange rate returns is expressed as
et Lj Ft ;
where for intervention days (j 1)
(4)
3 EUR1 t 6 7 6 INR1 7 t 7 6 6 7 6 JPY1 t 7 6 7 6 17 4 GBPt 5 SLR1 t
3 2 0 6 7 1 72 u 3 0 0 0 7 6 6 g1 g1 0 7 1;t 6 l0 l1 7 6 76 6 2 0 g1 2 7 6 76 u2;t 7 g 0 0 0 0 6 7 7 2 2 6 76 7 6 0 0 6 76 u 7 1 1 7wt 6 7 l l 6 g g 0 0 0 0 7 3;t 7: 3 7 6 3 3 3 6 76 0 6 6 7 7 6 7 1 u 6 0 4 7 g4 g4 0 0 0 0 6 7 4;t 5 6 l4 l1 7 4 7 4 0 5 u 6 7 5;t 5 4 g5 g1 0 0 0 0 5 1 l0 l 5 5

1 l0 1 l1
2
3
(5)
3.1.3. Variance decompositions Using the assumption that the factors are iid(0,1) random variables, Equations (2) and (3) are used to express the volatility of each of the currency returns into its component factors. For intervention days the total volatility of each currency return is
h 2 i 2 0 1 2 1 E e1 li li Var e1 g0 : i;t i gi
i;t
(6)
The proportion of the volatility of the return of exchange rate i when j 1 explained by the global factor wt, is

1 l0 i li
1 l0 i li
2 (7)
2
2 : g0 g1 i i
The proportion of the volatility of the return of exchange rate i, explained by the domestic factor ui,t, is
g0 g1 i i
2
2 2 : (8)
1 l0 i li
g0 g1 i i
On non-intervention days (j 0), the variance decompositions are the same, but with the structural break terms suppressed. 3.2. A factor model of central bank intervention To examine the effectiveness of central bank intervention, the model in Section 3.1 is extended by introducing intervention (net purchases of US dollars) as an endogenous variable. The dataset is again
34
j
separated into the two parts (j 0,1) of non-intervention vis--vis intervention days. Redening et to consist of n 6 series of zero mean bilateral exchange rate returns expressed against the US dollar and demeaned net intervention in millions of US dollars, the dataset is
et
EURt ; INRt ; JPYt ; GBPt ; SLRt ; INTt
j 0; 1:
(9)
The model for intervention in the Sri Lankan rupee exchange rate returns market rests on the assumption that intervention conducted by the Central Bank of Sri Lanka does not directly affect the exchange rate returns against the US dollar for the remaining exchange rates in the sample. Hence, the equations for the exchange rate returns for n 1,2.0.4 are the same as those stated in Equations (2) and (3). The new variable intervention INTj t ; is a function of the global factor wt with parameter loading j lint ; and an idiosyncratic factor vt with loading gjint such that when j 0
0 0 INT0 t lint wt gint vt ;
(10)
and when j 1
INT1 t
1 0 1 l0 int lint wt gint gint vt :
The endogenous treatment of intervention and its inclusion when j 0 provides a natural test of the model, as the variation in intervention is expected to be explained only mainly by its own idiosyncratic factor, with little effect from the global factors. The equation for the Sri Lankan rupee returns is the same as in Equation (2) for non-intervention days, but differs from Equation (3) for intervention days where
e1 5; t
1 0 1 1 l0 5 l5 wt g5 g5 u5;t sint vt :
(11)
The Sri Lankan rupee returns are now explained by the global factor wt, and two domestic factors. These are its own domestic factor u5,t and the domestic component of the intervention factor vt. On intervention days, the factor vt becomes an intervention factor, with the effectiveness of foreign exchange intervention by the central bank measured by the loading on the intervention factor in the Sri Lankan rupee exchange rate returns equation, s1 int : Only the domestic intervention factor is included in the Sri Lankan returns equation on intervention days as the global factor already affects both the Sri Lankan returns and Sri Lankan intervention decision. Hence the domestic intervention factor represents pure intervention as the effectiveness of this component of intervention is what the central bank controls outside of the impact of global shocks. In matrix form
et Lj Ft ;
and the model of central bank intervention is expressed as
(12) 3
2 2
3 3 6 2 0 7 2 3 g1 g1 0 0 0 0 0 6 0 1 7 EUR1 1 t u 6 l l 7 7 1;t 6 7 6 2 2 6 7 0 1 1 76 u 7 6 INRt 7 6 6 g2 g2 0 0 0 0 0 7 76 2;t 7 6 7 6 6 0 76 6 7 6 l0 l1 7 7 6 1 7 g g 76 u3;t 7 6 JPY1 7 6 0 0 0 0 0 3 3 6 7 3 3 t 76 6 7 6 7 6 w 7 t 76 u 7; 6 7 6 6 1 1 7 g0 0 0 0 0 0 76 4;t 7 6 GBP1 6 l0 4 g4 t 7 7 4 l4 6 7 6 7 6 7 6 0 7 74 u5;t 5 6 SLR1 7 6 6 6 0 1 7 g5 g1 s1 0 0 0 0 5 4 4 5 int t 5 6 7 0 6 l5 l5 7 vt gint g1 0 0 0 0 0 4 INT1 5 t int 0 1 lint lint (13)
when j 1.
1 l0 1 l1
35
3.2.1. Volatility decompositions Analogous to the factor model of exchange rate returns, the volatility decompositions for the factor model of central bank intervention is calculated using the expressions for the total variance for each type of variable
2 0 1 2 l l g1 ; i 1; 2; .; 4 Var e1 g0 i i i;t i i 2 0 1 2 1 2 s1 l5 l5 g0 ; Var e1 5;t 5 g5 intv 2 2 0 1 g1 lint lint g0 Var INT1 t int int :
(14)
For the Sri Lankan rupee returns, the percentage of the volatility of the returns explained by the global factor wt is

1 l0 i li
2
2 2 $100: 1 1 g s g0 i i intv
1 l0 i li
2 (15)
The percentage of the volatility of the returns explained by its own domestic factor u5,t is

1 l0 i li
2
2 2 $100: g0 g1 s1 i i intv
g0 g1 i i
2 (16)
Finally, the percentage of the volatility of the returns explained by the intervention factor vt is

1 l0 i li
2
g0 g1 i i
s1 int
2 2 2 $100: s1 intv (17)
In the same manner, the percentage of the volatility of intervention is decomposed into global (wt) and idiosyncratic (vt) factors, as

and
1 l0 int lint
2 (18)
1 l0 int lint
2
2 $100; g0 g1 int int
g0 g1 int int
2
2 (19)
1 l0 int lint
2 $100; 1 g g0 int int
respectively. 4. Estimation method The factor models of exchange rate returns and central bank intervention specied in the previous section use a GMM estimator, the estimates of which are known to be consistent, asymptotically normal and efcient (Hansen, 1982). GMM estimation does not require any extra information aside from that contained in the moment conditions. The estimation involves computing the unknown parameters by equating the theoretical moments of the model to the empirical moments of the data for both intervention and non-intervention day regimes in both models.
36
The model of intervention is identied and estimated by exploiting the feature of the data that intervention did not occur on some days and did occur on others.10 Inspection of Table 1 raises the possibility that the nature of the distributions on the two types of days (non-intervention versus intervention) across the low and global volatility period may change. The identication assumptions of the model also allow the underlying distributions on intervention and non-intervention days to differ as each regime uses the empirical moment conditions which are specic to the type of day to identify the model regime parameters, implicitly capturing any changes in the underlying joint distribution of the dataset on each day. In the case of the factor model of central bank intervention, which contains n 6 variables, there are a total of 42 moment conditions with which to identify 27 parameters by equating the empirical and theoretical moments of the model. Of the moment conditions, ((6 7)/2) 21 derive from nonintervention day data, with the additional 21 derived from intervention day data. The difference between the empirical moments and the theoretical moments of the model for each of the (non-intervention and intervention) regimes is
0 M0 vech U0 vech L0 L0 ;
and
(20)
0 M1 vech U1 vech L1 L1 ;
(21)
where Uj refers to the empirical variance-covariance matrices for non-intervention and intervention 0 days, and Lj Lj refers to the corresponding theoretical variance-covariance matrices for the two j regimes. The L derives from Equation (12) and uses the assumption that the factors are zero mean and unit variance, the empirical variance-covariance matrices are
Uj
1 X j j0 ee : Tj t T t t
j
(22)
where Tj represents the sample size of non-intervention and intervention day regimes.11 The objective function of the GMM estimator Q accounting jointly for both non-intervention days and intervention days is minimized according to
0 1 0 1 Q M0 W0 M0 M1 W1 M1 ;
(23)
where Wj are the optimal weighting matrices that correct for heteroskedasticity corresponding to j 0,1 (Hamilton, 1994; Newey and West, 1987). All calculations are undertaken using the library MAXLIK in GAUSS version 11. The GMM estimates are computed by iterating over the parameters and the weighting matrices using the BFGS algorithm with the gradients computed numerically. Note that in estimating the model, initial estimates of the variance-covariance matrices are obtained using identity weighting matrices. That is Wj I.12 When the number of moment conditions is greater than the dimensions of the parameter vectors, the model is over-identied. Empirically, an over-identication test can be used to check whether the models moment conditions match the data well or not. Using Hansens J-statistic, a test of the overidentication restrictions is given by
J TQ ;
(24)
10 Dungey et al. (2010) use the regimes of a non-crisis and a crisis period to identify models of contagion in much the same way that intervention is identied through the two regimes here of non-intervention days and intervention days. 11 Attempts to identify the model described in Section 3 using the variance-covariance matrices of the total dataset is infeasible, as this would generate only 21 empirical moments to identify 27 parameters in the theoretical model, leaving the model unidentied. 12 For some variants of the models, the use of optimal weighting matrices were infeasible; thus, for consistency, results using the identity-weighting matrix for all models are reported.
37
where T T0T1. The minimized distance given in Equation (24) is asymptotically distributed as a c2 with m b degrees of freedom, where m is the number of moment conditions and b is the number of parameters. A rejection of these restrictions indicates that some variables in the information set fail to satisfy the orthogonality conditions. 5. Foreign exchange intervention and volatility This section examines the effect of foreign exchange intervention by estimating the models outlined in Section 3. The rst set of models are estimated for the low volatility period. Before estimating the fully specied model of central bank intervention in Section 5.2, a model of exchange rate returns without a formal role for intervention is rst estimated in Section 5.1. The role of intervention is formally introduced in Sections 5.2 and 5.3. Section 5.2 evaluates the effectiveness of intervention in general, and Section 5.3 evaluates the differences in the effectiveness of intervention when the Central Bank of Sri Lanka intervenes by purchasing US dollars vis--vis when intervention occurs through sales. The model of central bank intervention distinguishing between purchases and sales is then rerun over the global volatility sample period in Section 5.4. Finally, Sections 5.5 and 5.6 presents some robustness analysis, looking at the effects of emerging market interdependence in the context of currency intervention, as well as the response of central bank intervention when accounting for time zones and lags in the intervention reaction function. 5.1. A factor model of exchange rate returns The results of the factor model of exchange rate returns outlined in Section 3.1 are presented in Table 3. To recap, this model does not formally model intervention, but it does separate the data into non-intervention days and intervention days. The factor model of exchange rate returns examines the contribution of the global and domestic factors to overall volatility in exchange rate returns for Sri Lanka (and the other currencies) on the two types of days. The discussion is framed in terms of the contribution of each factor to the volatility of each variable in percentage terms as shown in Equations (15)(19). The top panel of Table 3 provides the percentage contribution of the global and domestic factors to overall volatility on non-intervention days. The bottom panel provides the percentage contribution of the global and domestic factors to overall volatility on intervention days. The J-test for this model with 10 degrees of freedom is satised with a value of 13.481 and a p-value of 0.198. The results provide interesting insights into overall movements in currency markets during the two regimes. On days when there is no intervention, the Sri Lankan rupee returns are dominated by the domestic factor with almost 100 percent of volatility arising from purely domestic sources. Table 4 presents the parameter estimates for the factor model along with the p-values. On days when there is
Table 3 Volatility decomposition of the factor model of exchange rate returns. Notes: contribution of each factor to total volatility, in percent. The model is estimated over the period January 1, 2002June 29, 2007 (See Equations (7) and (8)). Factors Global Non-intervention days (j 0) EURt INRt YENt GBPt SLRt Intervention days (j 1) EURt INRt YENt GBPt SLRt 41.983 7.001 41.614 70.751 0.245 39.999 25.010 45.028 53.091 15.445 Domestic 58.017 92.999 58.386 29.249 99.755 60.001 74.990 54.972 46.909 84.555 Total 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
38
Table 4 Parameter estimates of the factor model of exchange rate returns. Notes: the model is estimated over the period January 1, 2002June 29, 2007 (see Equations (2) and (3)). p-values are in parentheses. Global factors Parameters Non-intervention days (j 0) l0 EURt 1 INRt JPYt GBPt SLRt Estimates 0.646 (0.000) 0.234 (0.000) 0.668 (0.000) 0.854 (0.000) 0.043 (0.460) 0.289 (0.914) 0.093 (0.381) 0.036 (0.878) 0.178 (0.358) 1.185 (0.000) Domestic factors Parameters Estimates 0.760 (0.000) 0.855 (0.000) 0.791 (0.000) 0.549 (0.000) 0.873 (0.000) 0.250 (0.627) 0.016 (0.975) 0.379 (0.577) 0.003 (0.986) 0.087 (0.889)
l0 2 l0 3 l0 4 l0 5
g0 1 g0 2 g0 3 g0 4 g0 5 g1 1 g1 2 g1 3 g1 4 g1 5
Intervention days (j 1) l1 EURt 1 l1 INRt 2 JPYt GBPt SLRt
l1 3 l1 4 l1 5
intervention, the volatility decomposition for Sri Lanka changes substantially. As shown in the bottom panel of Table 3, on intervention days, the global factor increases in importance from 0.2 percent to 15 percent. This suggests that Sri Lankan policy makers respond to global movements rather than domestic market movements when intervening in currency markets. Providing further support to this view is that, on non-intervention days, the only insignicant parameter is the global factor for the Sri 0 Lankan rupee returns l5 . Similarly, on intervention days, the only signicant structural break 1 parameter is the global factor for Sri Lanka l5 (see Table 3). The analysis in Section 5.2 provides further evidence on whether this is actually the case when intervention is formally introduced into the model. On both non-intervention and intervention days, the emerging economy of India is most similar to Sri Lanka, with 93 percent of its volatility a result of domestic factors on non-intervention days. On intervention days, the weight of the global factor is also larger for India, at 25 percent compared to 7 percent for non-intervention days. Global factors play a larger role for developed countries, with around 42 percent of volatility for the euro and yen returns, and 71 percent for the pound on nonintervention days. 5.2. A factor model of central bank intervention Estimating the factor model of central bank intervention, which adds an equation for intervention to the factor model of exchange rate returns provides a good overall t to the data, with the p-value of the J-test of 0.120. The inclusion of intervention does not change the volatility decomposition too dramatically as shown by comparing Table 5 and Table 3. The equations for the currency returns in the factor model of central bank intervention remain the same as those in the factor model of exchange rate returns for all currencies apart from the Sri Lankan rupee. Inspection of the second panel of the volatility decomposition in Table 5 shows that the central bank is able to inuence the volatility outcomes by 5.5 percent through intervention which is arguably a substantial magnitude given that the data is in terms of daily returns. Table 6 reports the results of Wald tests on the joint signicance of the intervention parameters and the intervention terms with all being statistically signicant. Comparing the results from the factor model of exchange rate returns (Table 3) to the model of central bank intervention (Table 5) for Sri Lanka shows that in both models, global factors contribute around 16 percent to Sri Lankan rupee return volatility. The intervention factor absorbs some of the volatility that is attributed to the domestic factor in the previous model. Note that all of the structural break parameters in the model are jointly signicant as shown in Table 6. The biggest difference in the results for the currency returns between the two models is in the contribution of the global factor on non-intervention days to the returns of India and Sri Lanka. For India, the contribution of the global factor rises from 7 percent to 18 percent; for Sri Lanka, it rises from 0.2 percent to 5.5 percent. These increases in magnitude suggest that the inclusion of Sri Lankan central
39
Table 5 Volatility decomposition of the factor model of Central Bank intervention. Contribution of each factor to total volatility, in percent. The model is estimated over the period January 1, 2002June 29, 2007 (see Equation (13)). Factors Global Non-intervention days (j 0) EURt 44.039 INRt 18.367 JPYt 41.741 GBPt 64.577 SLRt 5.455 INTt 0.010 Intervention days (j 1) EURt 37.796 INRt 30.425 JPYt 40.812 GBPt 48.183 SLRt 16.782 INTt 3.265 Domestic 55.961 81.633 58.259 35.423 94.545 99.990 62.204 69.575 59.188 51.817 77.697 96.735 Intervention 5.521 Total 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
bank intervention means that greater weight is placed on the emerging markets in the global factor in the model and alludes to similarities between currency movements and the factors driving them for India and Sri Lanka.
5.3. Purchases versus sales To further investigate the effectiveness of intervention, the intervention data are split into days when the Central Bank of Sri Lanka intervenes by purchasing US dollars, and days when intervention occurs through sales. The model for intervention days is written as
e 5; t
0 l0 5 l5 wt g5 g5 u5;t sint vt ;
(25)
where denotes days of US dollar purchases, and
e 5; t
0 l0 5 l5 wt g5 g5 u5;t sint vt ;
(26)
where denotes days of US dollar sales. Hence, the factor model is jointly estimated in three parts rather than two, and this model again satises the J-test with 25 degrees of freedom and a p-value of 0.921. Table 7 presents the volatility decomposition for the three regimes. The results clearly indicate that the central bank is more effective on days of US dollar purchases (sales of Sri Lankan rupee), with 11 percent of volatility in the Sri Lankan rupee returns being due to central bank intervention. In contrast, on days of sales of US dollars (purchases of Sri Lankan rupee), intervention is less effective and explains only 2 percent of volatility. This suggests that the Central Bank of Sri Lanka is more successful in
Table 6 Wald tests of intervention and structural breaks in the factor model of Central Bank intervention. The model is estimated over the period January 1, 2002June 29, 2007 (See Equation (13)). Hypothesis Joint intervention parameters g1 H0 : s1 int int 0 Joint idiosyncratic and intervention parameters s1 g1 H0 : g0 int int int 0 Joint structural break parameters 1 i 1; 2:::; 6 H0 : li g1 i 0; DOF 2 3 12 Test statistic 90.248 69.368 2270.097 p-value 0.000 0.000 0.000
40
Table 7 Volatility decomposition of the factor model of Central Bank intervention distinguishing between intervention through purchases of US Dollars and Sales of US Dollars. Notes: contribution of each factor to total volatility, in percent. The model is estimated over the period January 1, 2002June 29, 2007. Factors Global Non-intervention days (j 0) EURt 44.386 INRt 18.349 JPYt 41.547 GBPt 63.083 SLRt 5.698 INTt 0.000 Days of purchases (j ) EURt 28.033 INRt 41.976 JPYt 39.258 GBPt 39.272 SLRt 22.183 INTt 2.583 Days of sales (j ) EURt 36.253 28.499 INRt JPYt 41.472 GBPt 0.000 SLRt 17.225 INTt 4.104 Domestic 55.614 81.651 58.453 36.917 94.302 100.000 71.967 58.024 60.742 60.728 67.004 97.417 63.747 71.501 58.528 100.000 80.722 95.896 Intervention 10.813 2.053 Total 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
inuencing the exchange rate when the pressure in currency markets is to appreciate the Sri Lankan rupee. This result is also consistent with the Central Bank of Sri Lanka being focussed on achieving its medium-term target of accumulating international reserves in the low volatility period. The preliminary statistics in Table 1 further conrm this as intervention in the low volatility period occurs on days on which the mean return for the Sri Lankan rupee is positive, while the mean returns of the other currencies are negative. This implies that Sri Lanka is purchasing other currencies at a low value compared to their own on these days, hence accumulating reserves in a seemingly strategic manner. It is worth commenting on the changing role of the Indian rupee in this model. On days when intervention occurs, through either purchases or sales, the global factor affects Indian rupee returns by substantially more than on non-intervention days, again alluding to a possible common factor between India and Sri Lanka. 5.4. Intervention in the global volatility period The rst objective of the central bank is to contain excessive volatility in the exchange rate in the short run. This objective is examined in this section using the period of global volatility corresponding to the recent global nancial crisis. It is expected that increased volatility in currency markets leads to more intervention as monetary authorities move to curb some of the volatility. This is veried in Table 1 which presents statistics on intervention during the global volatility period. There are proportionately more days when intervention took place in the global volatility period, and the standard deviation of intervention is also higher. Notably, the number of intervention days through sales of US dollars is higher than the number of days of purchases, suggesting that the central bank aims to prevent excess currency market volatility arising from negative short-run shocks (or those placing pressure on the domestic currency to depreciate). The model in Section 5.3, which distinguishes the effects of intervention through the purchases and sales of US dollars, is estimated for the period July 2, 2007 to December 31, 2010. The results reinforce the suggestion that the Central Bank of Sri Lanka is successful in meeting its rst objective of containing excessive currency market volatility in the short run, particularly when pressure is for a rupee depreciation. Table 8 shows the volatility decomposition corresponding to this period, and it clearly
41
Table 8 Volatility decomposition of the factor model of Central Bank intervention distinguishing between intervention through purchases of US dollars and sales of US dollars. Notes: contribution to total volatility, in percent. The model is estimated over the period July 2, 2007December 31, 2010. Factors Global Non-intervention days (j 0) EURt 79.921 INRt 12.583 JPYt 3.641 GBPt 59.704 SLRt 3.898 INTt 4.093 Days of purchases (j ) EURt 65.314 INRt 29.451 JPYt 2.650 GBPt 51.721 SLRt 15.384 INTt 0.823 Days of sales (j ) EURt 58.055 24.616 INRt JPYt 2.042 GBPt 0.000 SLRt 4.170 INTt 3.399 Domestic 20.079 87.417 96.359 40.296 96.102 95.907 34.686 70.549 97.350 48.279 82.210 99.177 41.945 75.384 97.958 100.000 84.649 96.601 Intervention 2.406 11.181 Total 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
indicates that central bank intervention is more effective on days of US dollar sales than on days of purchases, with 11 percent of volatility in Sri Lankan rupee returns due to central bank intervention. In contrast, on days of purchases, intervention explains only 3 percent of total volatility. The results for the global volatility period are in contrast to the low volatility period, where purchases of US dollars are more effective than sales. Intervention during the global volatility period on days of purchases is consistent with the model for the low volatility period, which sees the global factor increase in importance for overall volatility compared with non-intervention days. On days of US dollar purchases the global factor does not change by much. Most of the effects of intervention are absorbed from the domestic factor which falls from around 96 percent of total volatility when there is no volatility to 85 percent of total volatility on days of purchases. The asymmetry in the effectiveness of US dollar purchases and US dollar sales of opposite magnitudes in the low and global volatility periods is worthwhile of further investigation, particularly as it is consistent with the dual objectives of the central bank. To further understand why this asymmetry emerges, variance equality tests on the exchange rate returns and intervention data are contained in Table 9. The tests are of the null hypothesis that the variance of currency return (or intervention) i on days of intervention through purchases is equal to the variance of currency return i on days of intervention through sales, against the alternative that the variances are different. The tests are conducted for both the low and global volatility periods. The results of Table 9 are striking and lend further support to the conclusions of the paper. The panel presenting the results for the low volatility period shows that the null hypothesis of equal variances of the variables on days of purchases versus sales is rejected for both the Indian and Sri Lankan rupee as well as for the intervention variable itself. There are no differences in the variances for the euro, yen or pound on the days of purchases compared to the days of sales. In stark contrast to the regional differences in the variances in the low volatility period, the variances of the Indian and Sri Lankan rupees are not signicantly different in the global volatility period on days of purchases compared to days of sales. In fact, the role of the developed markets in the high volatility period is emphasized as the null hypothesis of no difference in variance is rejected for the developed countries in the model only. This result is consistent with the switch in policy preference of the central bank in the global volatility period to reduce currency market volatility.
42
Table 9 Variance equality tests on the exchange rate returns (percent) and intervention data. The tests are the Bartlett test, the Levene test and the BrownForsythe test. Notes: the null hypothesis is that the variance of currency return i on the days of intervention through purchases is equal to the variance of currency return i on the days of intervention through sales, against that alternative that the variances on the days of purchases versus sales are different. The tests are conducted for the low and high volatility period. The low volatility period is from January 1, 2002June 29, 2007, and the global volatility period is from July 2, 2007 December 31, 2010. See Sokal and Rohlf (1995), Levene (1960) and Brown and Forsythe (1974a, 1974b). p-values are in parentheses. Low volatility period Bartlett Degrees of freedom EURt INRt JPYt GBPt SLRt INTt 1 1.171 19.018 0.253 0.095 2.860 9.833 (0.279) (0.000) (0.615) (0.757) (0.091) (0.002) Levene (1559) 0.706 (0.401) 7.162 (0.008) 0.067 (0.795) 0.257 (0.613) 8.837 (0.003) 7.747 (0.006) Brown-Forsythe (1559) 0.696 (0.405) 7.404 (0.007) 0.076 (0.782) 0.220 (0.639) 8.287 (0.004) 9.332 (0.002) High volatility period Bartlett 1 28.686 1.466 16.422 14.646 11.460 0.207 (0.000) (0.226) (0.000) (0.000) (0.001) (0.649) Levene (1518) 8.947 (0.003) 1.148 (0.285) 4.802 (0.029) 4.527 (0.034) 0.872 (0.351) 1.472 (0.226) Brown-Forsythe (1518) 8.959 (0.003) 1.005 (0.317) 4.793 (0.029) 4.453 (0.035) 0.069 (0.793) 0.986 (0.321)
Although this paper does not to focus in detail on changes to decompositions for the remaining countries in sample, it is interesting to glean an insight into the global volatility period and the dynamics of the currency markets during this time. The results differ markedly in terms of decompositions for most currencies in the model in the global volatility period, particularly for the euro and yen. The global factor now contributes 80 percent to the euro exchange rate volatility on non-intervention days, reecting that the euro US dollar relationship is a key source of volatility. Similarly, the yen is now completely driven by idiosyncratic factors (96 percent). An interesting result across all three models is that in both the low and high volatility period, the sterling exchange rate volatility is determined entirely by domestic factors on days that Sri Lanka intervenes by selling US dollars. In contrast, on the days of purchases, the contribution of the domestic factor is 60 percent for the low volatility period, and is 48 percent for the high volatility period. Factor models cannot provide a strong structural economic interpretation of the reason that a particular factor clusters around one source of variability, but it can provide some clues. As discussed in Section 2, inspection of the skewness statistics of Table 1 is indicative of the behavior of investors during the global volatility period, where skewness for the pound is positive on all types of days, in contrast to the negative sign for all other returns. The sterling is potentially acting as a safe haven currency in the global volatility period, explaining the loading on the domestic factor for the pound on days when the central bank is responding to the global economy and attempting to sell US dollars to restrain the volatility of the Sri Lankan rupee. 5.5. Emerging market interdependence and currency intervention Modeling the set of currency returns simultaneously raises some potential policy challenges that central banks particularly in emerging counties may need to consider. One consistent result that has come through the various model specications in Section 5 is the seemingly common movements in the impact of the factor structures for the Indian rupee and Sri Lankan rupee returns. Consistently, on days on which Sri Lanka intervenes, the global factor contribution to currency market volatility increases for both countries indicating either a common emerging market factor, or an increase in the role that developed countries have on the emerging country returns. This occurs in both the low and high volatility periods. To further investigate the interdependence between the emerging country currency markets both on days on which Sri Lanka intervenes in the currency market and on days in which it doesnt, the model in Equations (2) and (3) are reestimated to also account for a common factor to the emerging market economies. The model is hence re-specied for India and Sri Lanka so that their currency returns are also a function of an emerging market factor (dt) with loading ki, i 2,5. For the nonintervention days for India (i 2) and Sri Lanka (i 5) the equation is
0 0 e0 i;t li wt gi ui;t ki dt i 2; 5; 0
j 0;
(27)
43
Table 10 Volatility decomposition of the factor model of exchange rate returns with an emerging market factor. Notes: Contribution of each factor to total volatility, in percent. The model is estimated over the period January 1, 2002June 29, 2007 (see Equations (27) to (29)). Factors Global Non-intervention days (j 0) EURt 41.946 INRt 16.782 JPYt 45.523 GBPt 49.882 SLRt 4.371 INTt 0.038 Intervention days (j 1) EURt 41.311 INRt 23.919 JPYt 42.530 GBPt 59.647 SLRt 9.354 INTt 2.144 Domestic 58.054 69.222 54.477 50.118 75.923 99.962 58.689 30.989 57.470 40.353 67.722 97.856 Emerging Intervention 9.824 Total 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
13.996
19.706
45.092
13.100
and for the intervention days (j 1) for India the equation is
e1 2; t
1 0 1 0 1 l0 2 l2 wt g2 g2 u2;t k2 k2 dt ;
(28)
and for Sri Lanka
e1 5; t
1 0 1 1 0 l0 5 l5 wt g5 g5 u5;t sint vt k2 dt :
(29)
The results of the estimation of this model specication for the low volatility period with intervention are contained in Table 10. Comparison of Table 10 with Table 5 shows that there is evidence of an emerging market factor between India and Sri Lanka. Inspection of these tables shows rstly that on non-intervention days, the emerging market common factor contributes 14 percent to the volatility of the Indian Rupee, and 20 percent to the volatility of the Sri Lankan Rupee. Sri Lanka seems to be particularly affected by the India-Sri Lanka joint factor on the days of no intervention (j 0). The intervention day factor structure also presents evidence of an emerging market factor on these days. It is potentially important that this factor is comparatively more substantial for India than for Sri Lanka, with the emerging market factor now contributing 45 percent to the volatility of the Indian rupee on the days that Sri Lanka intervenes. Comparison of Table 10 with Table 5 shows that the factor structures are quite similar for the remaining returns across the model. The differences for India and Sri Lanka actually come mainly from a transfer of the factor loadings from the domestic factors to the emerging factor rather than from the global factor. However, it is still evident that Sri Lanka is intervening in response to global rather than emerging market or domestic factors, at least in the low volatility period.13 The reasons for emerging markets intervening are very different to those of developed countries and the frequency is also much greater. The results of this paper illustrate the lack of research undertaken in this area (see Menkhoff, 2012 for a survey of the limited literature) and raise pertinent questions about interdependence between emerging market currency returns when one country intervenes. It is the case that emerging markets operate often as competitors and trading partners
13 A similar model was also estimated for the high volatility period with the qualitative results of including an emerging market factor holding.
44
Table 11 Volatility decomposition of the factor model of exchange rate returns with an adjustment for time zones. Notes: contribution of each factor to total volatility, in percent. The model is estimated over the period January 1, 2002June 29, 2007 (see Equation (30)). Factors Global Non-intervention days (j 0) EURt 10.930 INRt 23.981 JPYt 12.379 GBPt 6.810 SLRt 41.932 INTt 5.015 Intervention days (j 1) EURt 28.651 INRt 0.415 JPYt 49.380 GBPt 0.261 SLRt 5.549 INTt 0.326 Domestic 89.070 76.019 87.621 93.190 58.068 94.985 71.349 99.585 50.620 99.739 89.548 99.674 Intervention 4.903 Total 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
to each other, and so it is likely that intervention in one market is either not independent from other emerging markets, or may impact on each other in ways which are not yet investigated. The results highlight the need of further investigation of cross market intervention effects for emerging markets in a multivariate model. 5.6. Central bank intervention, time zones and lags in the intervention reaction function This section presents the results to some sensitivity experiments to determine the robustness of the model to time zone changes and lags in the intervention reaction function. The ve currency markets considered operate in three major time zones. The European markets are open from 08:00 to 16:30 UMT, India and Sri Lanka are open from 03:45 to 10:00 UMT, and Japan is open from 0:00 to 6:00 UMT. As all markets are open for some period of time in the day that the Sri Lankan market is open, this paper chose to use the contemporaneous currency dataset. However, to determine whether the results are robust to the opening times of the markets, the markets which close after the Sri Lankan market are lagged one day in model. The robustness of the model to this feature is assessed by changing the structure of the moments used to calculate the parameters through the GMM procedure. The empirical variance covariance matrixes for j 0,1 in (22) are calculated using
Uj
1 X j j j 0 ei;t 1 ej ej ej m;t INTt m;t INTt i ; t 1 Tj 1 t T

j
i 1; 3; 4;
m 2; 5
(30)
instead, with the lagged exchange rates replacing the contemporaneous exchange rates.14 A second exercise was undertaken where we allow for the possibility that the central bank reacts to previous days returns when deciding to intervene. A second experiment is conducted where the empirical variance covariance matrixes for j 0,1 in (22) are calculated using:
Uj
1 X j j 0 ei;t 1 INTj ej INT t t i;t 1 Tj 1 t T

j
i 1; 2; .5:
(31)
for all currency returns rather than just those outside of the time zone.
14 Time zone issues relating to the currency pairs under investigation should not be too big of an issue in this model as it is effectively possible to trade currencies over the entire 24 hour of a day.
45
Table 12 Volatility decomposition of the factor model of exchange rate returns with an allowance for lags in the intervention reaction function. Notes: contribution of each factor to total volatility, in percent. The model is estimated over the period January 1, 2002June 29, 2007 (See Equation (31)). Factors Global Non-intervention days (j 0) EURt 27.495 INRt 2.501 JPYt 42.116 GBPt 0.426 SLRt 20.004 INTt 0.000 Intervention days (j 1) EURt 4.385 INRt 38.130 JPYt 3.791 GBPt 17.024 SLRt 34.916 INTt 5.340 Domestic 72.505 97.499 57.884 99.574 79.996 100.000 95.615 61.870 96.209 82.976 62.447 94.660 Intervention 2.637 Total 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
The results of the two robustness exercises are contained in Tables 11 and 12. Table 11 shows that the adjustment to the time zones does not qualitatively change the results substantially.15 In contrast, the results in Table 12 are a little different. The factor structure changes quite a lot particularly for the Indian rupee, the pound and the Sri Lankan rupee on the non-intervention days, and also for the euro and the yen on the intervention days. In comparison to other factor models in the literature, we would expect that the factor loadings for the global factors would be higher than what they are in Table 12 (see for example Dungey, 1999). The factor loadings in the benchmark models of this paper are more in line with those in the literature than those in this nal robustness experiment. 6. Conclusion Foreign exchange intervention by central banks in emerging economies has only been studied to a limited extent, and the effect of such intervention is not well understood. Using a unique dataset and a new modeling framework, this paper contributed to this literature by estimating a latent factor model of central bank intervention in a multi-country setting. The case study was for the emerging economy of Sri Lanka, whose intervention policy objectives are in the short run to contain excessive exchange rate volatility, and in the medium run to accumulate international reserves. The factor structure provided a convenient method of identifying sources of currency market volatility by decomposing the currency returns of Sri Lanka and Sri Lankas major trading partners into a set of factors that included global, domestic and intervention factors. The model was identied using information on the absence or presence of intervention on a particular day. The moments of the data on days of no intervention were used to estimate global and domestic factors. The moments of the data on days of intervention were used to estimate structural change to the factor structure on the days of intervention, as well as the effect of pure intervention by the central bank. The advantage of latent factors meant that observable variables did not need to be formally specied. The effectiveness of intervention was assessed over two periods: i) a period of relatively low volatility in global nancial markets, from January 2002 to June 2007; and ii) a period of high volatility (a global volatility period), corresponding to the global nancial crisis from July 2007 to December 2010. The results were directly linked to the objectives of the central bank listed above. The empirical results were supportive of intervention being effective in Sri Lanka over the two periods, albeit in different ways. The results during both periods showed that the Central Bank of Sri Lanka
15
We also experimented with using rolling two day averages or returns to control for time zones.
46
responded to global movements in currency markets when they intervened, rather than movements specic to the domestic foreign exchange market. This suggests that the central bank attempted to shield the domestic economy from externally sourced uctuations. In the low volatility period, 11 percent of total volatility was explained by intervention through purchases of US dollars, compared to two percent of volatility in the case of intervention through sales of US dollars. These ndings were consistent with the medium-term objective of the Central Bank of Sri Lanka of accumulating foreign exchange reserves, suggesting successful reserves management between 2002 and 2007. In contrast to the dominance of intervention through purchases relative to sales for the low volatility period, the central bank was focused on mitigating excess currency market volatility arising from short-run shocks during the global volatility period in the late part of the sample. The variance decompositions calculated for 2007 to 2010 clearly showed that eleven percent of Sri Lankan currency market volatility was explained by sales of US dollars as the central bank attempted to absorb some of the global turmoil in currency markets through exchange rate management. Finally, the results indicate that intervention of an emerging market may affect, and be affected by currency markets of neighboring emerging markets. The evidence here was for regional emerging market intervention effects, but the results point to the need for a broader investigation of the cross market effects of the emerging market intervention. References
Adler, G., Tovar, C.E., 2011. Foreign Exchange Intervention: a Shield against Appreciation Winds?. IMF Working Paper, WP/11/165. Aruman, S., 2003. The Effectiveness of Foreign Exchange Intervention in Australia: a Factor Model Approach with GARCH Specications. Queensland University of Technology School of Economics and Finance. Discussion Paper, No. 135. Board of Governors of the Federal Reserve System, 2007a. Press Release. August 7, 2007. http://www.federalreserve.gov/ newsevents/press/monetary/20070807a.htm. Board of Governors of the Federal Reserve System, 2007b. Press Release. August 10, 2007. http://www.federalreserve.gov/ newsevents/press/monetary/20070810a.htm. Board of Governors of the Federal Reserve System, 2007c. Press Release. June 28, 2007. http://www.federalreserve.gov/ newsevents/press/monetary/20070618a.htm. Brown, M.B., Forsythe, A.B., 1974a. Robust tests for the equality of variances. Journal of the American Statistical Association 69 (346), 364367. Brown, M.B., Forsythe, A.B., 1974b. The small sample behavior of some statistics which test the equality of several means. Technometrics 16 (1), 129132. Canales-Kriljenko, J., 2003. Foreign Exchange Intervention in Developing and Transition Economies: Results of a Survey. IMF Working Paper, WP/03/95. Central Bank of Sri Lanka, 2007. Exchange Rate. Central Bank of Sri Lanka. Pamphlet Series No.3. http://www.cbsl.gov.lk/ picsndocs/10pub/docs/pa/pamphlet/pl3.pdf. Retrieved June 2011. Diebold, F., Nerlove, M., 1989. The dynamics of exchange rate volatility: a multivariate latent factor ARCH model. Journal of Applied Econometrics 4 (1), 121. Disyatat, P., Galati, G., 2007. The effectiveness of foreign exchange intervention in emerging market countries: evidence from the Czech Koruna. Journal of International Money and Finance 26 (3), 383402. Dominguez, K., 2010. International reserves and underdeveloped capital markets. In: Reichlin, L., West, K. (Eds.), NBER International Seminar on Macroeconomics 2009. University of Chicago Press for the NBER. Dominguez, K., Hashimoto, Y., Ito, T., 2011. International Reserves and the Global Financial Crisis. NBER Working Paper, No.17362. Dungey, M., 1999. Decomposing exchange rate volatility around the Pacic Rim. Journal of Asian Economics 10 (4), 525535. Dungey, M., Fry, R., Gonzlez-Hermosillo, B., Martin, V.L., Tang, C., 2010. Are Financial Crises Alike?. IMF Working Paper, WP/10/14. Fry, R.A., Martin, V.L., Tang, C., 2010. A new class of tests of contagion with applications. Journal of Business and Economic Statistics 28, 423437. Guimares, R., Karacadag, C., 2004. The Empirics of Foreign Exchange Intervention in Emerging Market Countries: the Cases of Mexico and Turkey. IMF Working Paper, WP/04/123. Hamilton, J., 1994. Time Series Analysis. Cambridge University Press. Hansen, B., 2000. Testing for structural change in conditional models. Journal of Econometrics 97 (1), 93115. Hansen, L., 1982. Large sample properties of generalized method of moments estimators. Econometrica 50, 10291054. Herrera, A., Ozbay, P., 2005. A Dynamic Model of Central Bank Intervention. Central Bank of the Republic of Turkey. Working Paper, No. 0501. Harvey, C.R., Siddique, A., 2000. Conditional skewness in asset pricing tests. The Journal of Finance 55, 12631295. Kamil, H., 2008. Is Central Bank Intervention Effective under Ination Targeting Regimes? the Case of Colombia. IMF Working Paper, WP/08/88. Levene, H., 1960. Robust tests for equality of variances. In: Contributions to Probability and Statistics: Essays in Honor of Harold Hotelling, vol. 2, p. 278. Mahieu, R., Schotman, P., 1994. Neglected common factors in exchange rate volatility. Journal of Empirical Finance 1, 279311. Mandeng, O., 2003. Central Bank Foreign Exchange Market Intervention and Option Contract Specication: the Case of Colombia. IMF Working Paper, WP/03/135.
47
Menkhhof, L., 2012. Foreign Exchange Intervention in Emerging Markets: a Survey of Empirical Studies. Discussion Paper No. 498. Department of Economics, Leibniz Universitt Hannover. Mulder, C., Perrelli, R., 2001. Foreign Currency Credit Ratings for Emerging Market Economies. IMF Working Paper, WP/01/191. Newey, W., West, K., 1987. A simple, positive semi-denite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica 55, 703708. Pattanaik, S., Sahoo, S., 2001. The Effectiveness of Intervention in India: an Empirical Assessment. Reserve Bank of India. Occasional Papers, No. 22. Pointines, V., Rajan, R.S., 2011. Foreign exchange market intervention and reserve accumulation in emerging Asia: is there evidence of fear of appreciation? Economics Letters 111 (3), 252255. Sarno, L., Taylor, M., 2001. Ofcial intervention in the foreign exchange market: is it effective and, if so, how does it work? Journal of Economic Literature 39, 839868. Sokal, R.R., Rohlf, F.J., 1995. The Principles and Practice of Statistics in Biological Research, New York, ed., 3. Szakmary, A.C., Mathur, I., 1997. Central bank intervention and trading rule prots in foreign exchange markets. Journal of International Money and Finance 16 (4), 513535.

Currency Intervention A Case Study of An Emerging Market

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Currency Intervention A Case Study of An Emerging Market

Transféré par

Droits d'auteur :

Formats disponibles

Journal of International Money and Finance 37 (2013) 2547

Contents lists available at SciVerse ScienceDirect

Journal of International Money and Finance

Currency intervention: A case study of an emerging marketq

Indian rupee returns

2009 2009 2009

Japanese yen returns

British pound returns

120 80 40 0 -40 -80 -120 -160

120 80 40 0 -40 -80 -120 -160

EURt ; INRt ; JPYt ; GBPt ; SLRt

3 EUR1 t 6 7 6 INR1 7 t 7 6 6 7 6 JPY1 t 7 6 7 6 17 4 GBPt 5 SLR1 t

EURt ; INRt ; JPYt ; GBPt ; SLRt ; INTt

1 0 1 l0 int lint wt gint gint vt :

2 2 2 $100:  s1 intv (17)

2 $100;  g0 g1 int int

2 $100;  1 g g0 int int

Intervention days (j 1) l1 EURt 1 l1 INRt 2 JPYt GBPt SLRt

where denotes days of US dollar purchases, and

and for the intervention days (j 1) for India the equation is

and for Sri Lanka

  1 X j j j 0 ei;t 1 ej ej ej m;t INTt m;t INTt i ; t 1 Tj 1 t T

  1 X j j 0 ei;t 1 INTj ej INT t t i;t 1 Tj 1 t T

Vous aimerez peut-être aussi

2 2 2 $100: s1 intv (17)

2 $100; g0 g1 int int

2 $100; 1 g g0 int int

1 X j j j 0 ei;t 1 ej ej ej m;t INTt m;t INTt i ; t 1 Tj 1 t T

1 X j j 0 ei;t 1 INTj ej INT t t i;t 1 Tj 1 t T