# An exchange market pressure measure for cross-country

## Comments

## Transcription

An exchange market pressure measure for cross-country

NIPFP Working NIPFP Working paper paper series series An exchange market pressure measure for cross-country analysis No. 189 28-Feb-2017 Ila Patnaik, Joshua Felman and Ajay Shah National Institute of Public Finance and Policy New Delhi Working paper No. 189 An exchange market pressure measure for cross country analysis Ila Patnaik National Institute of Public Finance and Policy (NIPFP), New Delhi Joshua Felman NIPFP, New Delhi Ajay Shah1 NIPFP, New Delhi Abstract EMP measures in the existing literature are oriented towards applications in crisis dating and prediction. We propose a modified EMP measure where cross-country comparisons are possible. This is the sum of the observed change in the exchange rate with an estimated counterfactual of the magnitude of the change in the exchange rate associated with the observed currency intervention. We construct a multi-country dataset for EMP in each month. This opens up many new research possibilities. JEL Codes: E52, F31, F32 Keywords: Exchange rate regime, capital flows, currency wars, monetary policy, Exchange market pressure, Statistical system. Email addresses: [email protected] (Ila Patnaik), [email protected] (Joshua Felman), [email protected] (Ajay Shah) 1 The authors would like to thank Michael Hutchison, Philip Lane, Tarun Ramdorai and Rex Ghosh, seminar participants at the 2013 LACEA Annual meetings, the NIPFP DEA Research meetings, the International Monetary Fund, Trinity College Dublin and UC Santa Cruz for helpful comments and suggestions. We would like to thank Shekhar Hari Kumar, Vimal Balasubramaniam and Mohit Desai for excellent research assistance. The views expressed in this paper are those of the authors and do not necessarily reflect those of the institutions. Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 2 Working paper No. 189 1. Introduction The concept of exchange market pressure was first proposed by Girton and Roper (1977). The notion is straight forward: emp measures the total pressure on an exchange rate, which has been resisted through foreign exchange intervention or relieved through exchange rate change. The problem is that measuring emp requires combining the observed change in the exchange rate, which is a percentage change, with an observed intervention which is measured in dollars. The early efforts in measuring emp worked directly using monetary models, whereas more recent efforts have focussed on measuring emp using indices that combine changes in reserves and the exchange rate. The direct measures of emp are model dependent and primarily geared towards finding the magnitude of money market disequilibrium that must be removed either through reserve or exchange rate changes under any desired exchange rate target. emp indices meanwhile, are designed to capture and forecast crises. Direct measures often lack consistent units, while indices do not have this problem and are better suited to crisis conditions. Girton and Roper (1977) assumed that foreign exchange intervention was unsterilised. Thus intervention led to equivalent amounts of changes in base money. Money was assumed to be neutral so that percentage changes in base money led to equivalent changes in prices. The assumption of purchasing power parity meant that percentage changes in domestic prices were essentially equal to exchange rate changes. Under these assumptions, the authors added the percentage changes in reserves and in exchange rates. However, monetary models had low predictive power for changes in exchange rates and the resulting measure of emp was often misleading (Eichengreen et al., 1996). When emp is measured as : EGR = ∆et + ∆r̄t the first right-hand side term is in the units of percentage change of the exchange rate, and the second is in the units of percentage change of reserves as a fraction of monetary base. This formula could only been motivated by the assumption that for all countries, at all time periods, a reserves change of 1% of m0 (monetary base) has an impact on the currency of 1%. But there is no basis for expecting the foreign exchange market to have such a property for all countries and for all times. In order to address these problems, Eichengreen et al. (1996) created a new measure of emp. They normalised all prices and quantities, then weighted these components of the index by the inverse of their historical volatilities. Alternative weighing schemes were proposed by Sachs et al. Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 3 Working paper No. 189 (1996); Kaminsky et al. (1998); Pentecost et al. (2001); Klaassen (2011); IMF (2007). This approach has led to many useful and important applications in international finance and macroeconomics. The emp indices, however, have well documented problems with the “arbitrary” choice of index weights and crisis thresholds (Pontines and Siregar, 2008). In addition, normalisation means the emp indices can not be used for cross-country comparisons; they are designed for comparison across time series of a country to indicate periods of “extreme” emp. Under a fixed exchange rate, many of the conventional measures yield an emp of infinity, which hampers applications. Consider a research question such as the impact of quantitative easing (QE) upon emerging markets. It is natural to look at this common shock (QE) inducing exchange market pressure upon all EMs. An array of questions can then be asked. What were the country characteristics which led to high EMP in some emerging markets (EMs) but low EMP in others? Which EMs allowed EMP to be expressed as exchange rate fluctuations, and which EMs did not? What were the causes and consequences of fear of floating? These questions require measurement of EMP in a way that permits comparisons across countries and time. Consider a practical question such as the outcome of the US presidential election in 2016. It would be useful to observe the EMP and exchange rate changes across all countries of the world in November and December 2016. This could be a useful tool for finance practitioners and for policy makers. This also requires measurement of EMP in a way that permits comparisons across countries and time. Towards this objective, we build on Weymark (1995), who added the change in the exchange rate that was observed with that of the change in the exchange rate that was prevented by the central bank through intervention or by changes in the policy rate. This measure has a consistent unit: the percent change in exchange rate over a one-month period. This takes us to the question: What is the magnitude of the exchange rate movement associated with $1 billion of intervention? There are many problems in estimating this. Intervention and the exchange rate level may be endogenous. The impact of foreign exchange intervention, when there is any impact at all, may be asymmetric depending on the direction of the intervention, time varying and temporary (Menkhoff, 2013; Disyatat and Galati, 2007; Lahura and Vega, 2013). We turn to concepts from the working of financial markets, where the notion of ‘market impact’ is used when understanding the price change associated with large trades placed by investors. The impact of a billion dollars of intervention depends on the size and liquidity of the currency market. Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 4 Working paper No. 189 We propose an estimation strategy through which the exchange rate change associated with $1 billion of intervention is measured for some countries for some points in time. Our method relies on situations where a country has switched between a fixed and a floating exchange rate regime (or vice versa). Assuming similarity of macroeconomic shocks before and after the change in the exchange rate regime, we are able to obtain an estimate of the exchange rate change associated with $1 billion of currency intervention. In our analysis of the global data, we find 39 country-periods where such estimation is possible. Regression analysis of these values is used to impute values for other country-year settings. This gives the ability to measure emp for all countries for all months, in a way that is comparable across countries and months. We apply a series of sanity checks and robustness checks, and find that this database has meaningful properties. This is a paper focused on measurement. It results in a dataset with information about monthly EMP for a large panel of countries. The dataset is released in the public domain and is regularly updated by the authors. An array of interesting research questions, and real world applications, could flow from this work. 2. Measures of EMP An emp index consists of a sum of a standardised change in the exchange rate and a standardised change in reserves, both of which are dimensionless and hence conformable for addition. emp indices were developed for the purpose of analysing currency crises, one country at a time. Crisis periods, in general, are periods when policy makers were often trying to defend the exchange rate, using all possible policy options. All components of emp are generally seen to move up in this period. When each of these is first standardised, and then added up to obtain an index, the index has high values for periods of crisis, where high is often identified as the index being some standard deviations away from the norm. These indices are, however, not appropriate for cross-country comparisons. We define It as the intervention of the central bank in time t. The exchange rate is denoted by et , reserves by rt , base money as m0 and reserves divided by base money by r̄t . The change in et is denoted by ∆et ; the change in rt is denoted by ∆rt . The change in mrt0 is denoted by ∆r̄t . Under this notation, some of the existing emp measures are: Eet (Eichengreen et al., 1996): Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 5 Working paper No. 189 1 ∆et 1 EMPt = − σe e t σr̄ ∆r̄t ∆r̄U St − r̄t r̄U St ∆et − et + 1 (∆ (it − iU St )) σi (1) (2) Est (Sachs et al., 1996): Est = Kt = 1 1 1 + + σe σr σi 1 σe Kt 1 σr Kt ∆rt + rt 1 σi Kt ∆it (3) Ekt (Kaminsky et al., 1998): Ekt ∆et = − et σe ∆rt σ r rt + σe ∆it σi (4) and Ept (Pentecost et al., 2001): an index based on a principal components analysis of the sub-components in Eichengreen et al. (1996). Eimf (IMF, 2007): EMP = 1 σ∆%ei,t ∆%ei,t + ∆%resi,t = ∆ei,t = 1 σ∆%resi,t ∆%resi,t NFAi,t − NFAi,t−1 Monetary basei,t−1 eri,t − eri,t−1 eri,t−1 (5) (6) (7) In all these cases, for a fixed exchange rate regime, the standard deviation of the exchange rate which is the term in the denominator, is zero. This results in giving an infinitely large weight to the coefficient of exchange rate movements. Consequently, when a country with a pegged exchange rate allows small changes in the exchange rate to occur, these show up as a high emp because of the large weight being given to exchange rate changes. As an example, consider a historically inflexible exchange rate like that of China, where for long periods of time σ∆e ≈ 0. In periods when a small exchange rate change takes place, and the numerator is non-zero, a very large and spurious value for emp will be induced. Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 6 Working paper No. 189 Figure 1 emp index for China and the foreign exchange reserves build up 5 0 EMP Observed exchange rate change −5 % change in exchange rate This panel juxtaposes the emp index calculated for China, with observed exchange rate change and forex reserves as a percentage of base money. The emp index indicates small/zero appreciation pressure on the renminbi from 2003 to 2006. However, the massive reserve build up during the same period seems to suggest the renminbi was under strong pressure to appreciate. 2003 2004 2005 2006 2007 2008 2009 120 100 80 60 Reserves as % of base money Forex reserves as % of base money 2003 2004 2005 2006 2007 2008 2009 2010 We see the consequence of the large weight given to exchange rate movements, and the small weight given to intervention due to the large variation in intervention in the emp measure shown in Figure 12 . The emp appears low in periods when there was a large change in reserves, and higher when there was a small change in the exchange rate. Conversely, for floating exchange rate regimes, these measures give a large weight to intervention. In addition, there are measurement issues, as not all countries release intervention data. When reserve changes are used to 2 We use the emp measure given in equation 1 above, that is, as defined in Eichengreen et al. (1996). This measure will be used for all subsequent references to emp index. Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 7 Working paper No. 189 Figure 2 The emp index for four countries This panel shows the emp index for China, India, Brazil and Egypt. The magnitude of the emp appears similar across the four countries. Using this measure, it is not possible to tell whether India witnessed a different magnitude of pressure than China in the 2000s. 10 −10 0 Index 10 0 −10 Index 20 India 20 China 2004 2005 2006 2007 2008 2009 2004 2005 2008 2009 2008 2009 20 10 −10 0 Index 10 0 −10 Index 2007 Egypt 20 Brazil 2006 2004 2005 2006 2007 2008 2009 2004 2005 2006 2007 approximate intervention in countries where the exchange float is relatively clean, revaluation effects and interest income end up being given a large weight due to the low variance of reserves. These show up as large emp, spuriously signaling heavy exchange market pressure. These characteristics make conventional emp indices unsuitable for comparisons across countries. As an illustration Figure 2 shows the emp index for 4 countries: China, India, Brazil and Egypt. Each country’s emp depends on its historical experience as the measure uses standard deviation from historical data. As a consequence, there are no visible differences between the emp for a country like China that witnessed large appreciation pressure during the 2000s and the others that witnessed smaller exchange market pressure in both directions. The usefulness of other emp indices for cross country comparisons varies, but the essential argument for not using them for such comparisons remains unchanged. Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 8 Working paper No. 189 2.1. A new EMP measure In order to do cross country comparisons as well as comparisons across time, we propose an emp measure with consistent units - percent change in the exchange rate. The proposed measure adds the change in the exchange rate that took place, and the change that we expect would have occurred had there been no intervention. Both components are measured in the same units, i.e in terms of the percentage change in the exchange rate. To transform the intervention into a measure of the percentage change that was prevented, we need a conversion factor, which we denote as ρt . The challenge is that as conditions in foreign exchange markets evolve, the impact of intervention may vary. ρt is not a constant and may be expected to vary over time, and across countries. We propose to measure emp in units of percentage exchange rate change over a one-month period: empt = ∆et + ρt It • ∆et is the percentage change in the exchange rate, • It is the intervention measured in billion dollars, • ρt is the conversion factor, which is the change in the exchange rate associated with $1 billion of intervention. The value of the conversion factor will depend on size and liquidity of the foreign exchange market. It follows that ρt It is expressed in units of percentage change of the exchange rate. It is the exchange rate change of the month we would have expected if there had been no intervention. A key question is whether the conversion factor can be estimated sufficiently well to produce a robust measure of emp. In this paper, we propose an empirical strategy for measuring ρt . We show that these values are consistent with our priors about what ρt ought to be. We go on to utilise these values to construct an emp database, which has attractive properties. 2.2. Estimating ρt Estimates of the impact of intervention in the literature vary highly due to identification problems. The impact depends on other policies such as sterilisation, communication or inflation targeting by the central bank (Menkhoff, 2013). For example, Evans and Lyons (2006) estimate the impact that ordinary order flow has on the exchange rate as 0.44 basis points per 10 million US dollar order flow in the highly liquid Deutsche Mark-US Dollar market in 1996. Scalia (2008) estimates an impact between 7 to 12 basis points per Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 9 Working paper No. 189 10 million euro for the Czech Republic. Tapia and Tokman (2004) estimate that in Chile, sales of US dollar in 1998-99 resulted in a 1 per cent exchange rate change on 500 million US dollar intervention. Guimares and Karacadag (2004) estimate that 100 million US dollar sales has an impact on the Mexican peso of 0.4 per cent, whereas purchases have no effect. Though the estimates are not strictly comparable, ρ estimates in this literature, when translated into our framework, lie between 0 and 10 per cent impact upon the exchange rate, of a billion dollars of intervention. While these papers are useful for obtaining an intuitive sense of the plausible magnitudes, for the purpose of a data-driven algorithm that utilises cross-country data to create a panel database about emp, we require an estimation strategy which yields estimates of ρt on a global scale. We propose going about this in two steps. The first step is to estimate ρt in certain situations in the data. The second step is to find the determinants of the estimated ρt and to use these to predict ρt for all country periods and years. The first step is based on a key insight which yields an identification opportunity. Assume a country which has experienced both fixed and float periods. Assume during the fixed periods, the country only uses intervention to influence the exchange rate, and that it does no intervention during the float periods. These are highly restrictive assumptions, but necessary to permit identification of ρ (We return to this issue in Section 3). Accordingly, we observe ∆et in float periods and It in fixed periods. EMPt = ∆et + ρt It EMPfloat = ∆et EMPfixed = ρt It In order to identify “normal times” which do not have unusual macroeconomic volatility, we exclude countries with currency crises. In normal times, we argue that macroeconomic shocks and hence emp volatility are similar across these periods and consequently, emp volatility. Under this assumption: V ar(EMPfixed ) = V ar(EMPfloat ) ρt = V ar(∆efloat ) V ar(Ifixed ) 1 2 (8) (9) This gives an opportunity for measuring ρt in some situations. To estimate ρ, we need to observe countries which have experienced both fixed and floating Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 10 Working paper No. 189 exchange rate regimes. These should be periods in which we can assume that the volatility of the exchange market pressure is roughly constant. The fixed and float regimes should be adjacent so that this is a relatively short window of time. We analyse 137 countries from Feb 1995 to Dec 2009, using the Zeileis et al. (2010) methodology to identify structural breaks in the de facto exchange rate regime. This methodology finds dates of structural change in the Frankel and Wei regression (Frankel and Wei, 1994). The R2 of the Frankel-Wei regression is our measure of exchange rate flexibility. For this purpose, we define a fixed exchange rate regime as a period when R2 > 0.95, and a floating exchange rate regime when R2 < 0.66. Each period is required to be at least 12 months long. The dates for structural change of the exchange rate regime are validated against the Reinhart and Rogoff (2004) exchange rate regime breaks (Appendix 6.1). We exclude periods where macroeconomic shocks were known to be high in one of the periods and known crisis dates. We also remove periods defined as “freely falling” by Reinhart and Rogoff (2004), when the volatility of the exchange market pressure cannot be assumed to be constant. This gives us 26 events where a country moved from a floating to a fixed exchange rate regime, and 13 events where a country moved from a fixed to a floating exchange rate regime.3 We estimate ρ using the above methodology for each of the 39 regime break points in our dataset for which such assumptions can be made. Table 1 show the estimated values of ρt associated with episodes of break dates that involve a movement from a floating exchange rate regime to a fixed exchange rate regime. Table 2 shows the values of ρt s estimated when countries move from a fixed exchange rate regime to a floating regime. For every episode, the value of ρt is attributed to the mid-point of the window for estimation. As an illustration, we show all the steps involved in estimating ρt for one example: Kenya. This is one of the countries seen in Table 1 which moved from a floating rate to a fixed rate. Figure 3 shows the dates of structural break of the exchange rate regime. From April 1997 to July 2001, the Kenyan shilling was floating. This is followed by a period from July 2001 till December 2002 when the Kenyan Shilling was pegged to the USD and the Kenyan central bank was intervening in the currency market. Reinhart and Rogoff (2004) identify this entire period as a de-facto crawling peg regime: this highlights the improvements in exchange rate regime analysis obtaind using the ZSP methodology. Using equations 8 and 9, our estimation of ρt is 105 percent per billion dollars. This suggests that a million dollars of intervention by the Central Bank of Kenya in currency markets would have prevented a 0.105% change in the exchange rate in the period July 2001 to December 2002. The number makes intuitive sense when compared with other estimates of the impact of intervention. Figure 4 shows an estimate of the change in the exchange rate that would have 3 Section 6 (Appendix) shows that periods identified by us as float roughly match the Reinhart and Rogoff (2004) classification of managed float. Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 11 Working paper No. 189 Table 1 Episodes of transition from float to fixed This table shows 26 structural change events for currency regime from float to fix. These country-periods have been used to estimate ρ. As an example, Angola switched from a float to a fix in May 2007, and this episode yields an estimate of ρt = 3.08; i.e. a billion dollars of intervention would yield a 3.08% change in the exchange rate. Country Angola Bangladesh Brazil Belarus Cape Verde Djibouti Djibouti Ethiopia Guinea Guyana India Kenya Comoros Kazakhstan Laos Sri Lanka Mongolia Maldives Malaysia Tunisia Trinidad and Tobago Trinidad and Tobago Ukraine Venezuela Vietnam Antigua and Barbuda Float period Nov 2006-May 2007 Dec 2005-Jan 2007 Jun 1994-Jul 1995 Jun 2009-Apr 2010 Mar 1999-Sep 2001 Jun 1996-May 1997 Mar 2002-Oct 2002 Sep 2002-May 2007 Aug 1998-Sep 1999 Oct 1998-Jul 1999 Aug 1997-Aug 1998 Apr 1997-Jul 2001 Jul 2004-May 2006 Mar 2006-Sep 2007 Jun 2001-Nov 2001 Jun 2000-Jun 2001 Sep 1998-Mar 2001 May 2005-Apr 2006 Aug 1997-Aug 1998 Sep 1990-Sep 1991 Sep 1996-Oct 1997 May 2008-May 2009 Mar 2008-Nov 2009 Feb 2002-Sep 2003 Sep 2000-May 2001 Feb 1996-Aug 2002 Accessed at http://www.nipfp.org.in/publications/working-papers/1779 R2 0.55 0.62 0.51 0.59 0.31 0.34 0.53 0.65 0.55 0.48 0.50 0.54 0.48 0.58 0.44 0.48 0.45 0.46 0.21 0.47 0.59 0.58 0.18 0.34 0.66 0.63 Fix period May 2007-Feb 2009 Jan 2007-Oct 2010 Jul 1995-Jan 1999 Apr 2010-Apr 2011 Sep 2001-Oct 2002 May 1997-Dec 1999 Oct 2002-Jul 2004 May 2007-Jan 2009 Sep 1999-Aug 2001 Jul 1999-Jun 2005 Aug 1998-Mar 2004 Jul 2001-Dec 2002 May 2006-Dec 2006 Sep 2007-May 2011 Nov 2001-Oct 2003 Jun 2001-Apr 2002 Mar 2001-Dec 2001 Apr 2006-Jan 2007 Aug 1998-Jul 2005 Sep 1991-Aug 1992 Oct 1997-Jun 1999 May 2009-Sep 2010 Nov 2009-Dec 2011 Sep 2003-Jan 2010 May 2001-Mar 2008 Aug 2002-Oct 2011 R2 0.99 0.95 0.99 0.97 1.00 0.99 1.00 0.96 1.00 0.99 0.97 0.97 0.96 0.99 1.00 0.95 0.96 0.96 1.00 0.99 0.99 0.96 0.99 1.00 1.00 1.00 ρt 3.08 6.85 1.97 5.41 669.26 604.14 376.19 8.68 268.97 442.47 1.55 105.64 462.02 2.48 390.10 28.20 184.94 79.04 5.35 38.69 19.18 7.52 6.49 8.82 1.07 36.59 Page 12 Working paper No. 189 Table 2 Episodes of transition from fixed to float This table shows 13 fixed to float country-periods which have been used to estimate ρ along with the ρ estimates for those periods. The R2 in the fixed periods are very high and those in float periods much lower. As an example, Costa Rica switched from fixed to float in January 1997, which yields an estimate of ρt = 5.83; i.e. a billion dollars of intervention would yield a 5.83% change in the exchange rate. Country Costa Rica Cape Verde Djibouti Gambia Guyana Laos Moldova Mauritius Malaysia Tunisia Ukraine Vietnam C African Republic Fix period Mar 1996-Jan 1997 May 2003-Jul 2004 Dec 1995-Jun 1996 Jul 1997-Dec 1998 Jul 1999-Jun 2005 Apr 2000-Jun 2001 Apr 2000-Nov 2000 Apr 2001-Dec 2002 Nov 1989-Dec 1993 Sep 1991-Aug 1992 Aug 2002-Apr 2003 Nov 1997-Sep 2000 Jun 2001-May 2002 R2 0.99 0.99 1.00 0.95 0.99 1.00 0.95 0.98 0.96 0.99 1.00 1.00 0.99 Float period Jan 1997-Jul 1997 Jul 2004-Dec 2007 Jun 1996-May 1997 Dec 1998-Nov 2003 Jun 2005-Dec 2005 Jun 2001-Nov 2001 Nov 2000-May 2001 Dec 2002-May 2004 Dec 1993-Jul 1994 Aug 1992-Jan 1994 Apr 2003-Feb 2004 Sep 2000-May 2001 May 2002-Jan 2004 R2 0.41 0.44 0.34 0.50 0.49 0.44 0.56 0.62 0.44 0.61 0.59 0.66 0.50 ρt 5.83 343.33 206.08 691.35 269.32 519.76 208.57 117.42 2.58 25.64 15.74 6.41 520.58 Figure 3 Exchange rate regimes in Kenya The graph shows the full history of the Kenyan exchange rate regime. In this, Zeileis et al. (2010) classifies the period from April 1997 to July 2001 as a float with an R2 of 0.54, and the period from July 2001 to December 2002 as a fixed exchange rate regime with an R2 of 0.97. 0.97 80 60 40 KES/USD exchange rate 100 0.54 1995 Period 1 Period 2 2000 Start date 1997-04-11 2001-07-20 2005 End date 2001-07-20 2002-12-27 Accessed at http://www.nipfp.org.in/publications/working-papers/1779 2010 R2 0.54 0.97 Page 13 Working paper No. 189 Figure 4 Kenya: Exchange Market Pressure The figure presents an estimate of the change prevented in the exchange rate by intervention by the Kenyan central bank when the regime shifted from a float to a fix between July 2001 and Dec 2002. This suggests that without intervention, we may have observed greater volatility in exchange rate returns during this period. 2001−07−01 −5 0 5 % change per month 10 Observed Counter−factual 1998 1999 2000 2001 2002 2003 occurred had the central bank of Kenya not intervened in the currency market. This provides a measure of the exchange market pressure in the fixed period. In the floating period, emp can be seen as the observed change in the exchange rate. 2.3. Predicting ρt The conversion factor ρt is primarily about the liquidity of the currency market. The impact of central bank intervention on the foreign exchange market will vary by country, by time. As the size of a currency market changes, ρ will change. We therefore need to estimate a ρt time-series for each country to measure emp. Data for size of the foreign exchange market, in terms of the daily dollar turnover in the spot and derivatives markets is available for some countries and years from the Bank for International Settlement 4 . The numerical magnitude of ρ will tend to be smaller when the currency market is more liquid, i.e. for bigger and more internationalised countries with greater financial development.5 Our estimates of ρ for larger emerging markets like Brazil, Turkey, India, Malaysia, Belarus, Indonesia indeed show ρ in the range of 1 to 10, consistent with the literature (Section 2.2). Meanwhile, countries with very small 4 BIS Triennial Central Bank Survey of Foreign Exchange and Derivatives Market Activity 5 Klaassen and Jager (2011) and BIS (1993) note that the extent of intervention depends on the turnover in the foreign exchange market Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 14 Working paper No. 189 Table 3 Estimated ρt and foreign exchange market turnover This table shows examples of estimated ρ and foreign exchange market turnover. The evidence points towards a negative relationship between currency market turnover and impact of intervention Country Year ρt FX market daily turnover (in Billion USD) Brazil 1997 1.97 5 India 2001 1.55 3 Malaysia 2002 5.35 1 Source: BIS, Brazil data is for 1998 and Malaysia for 2001 economies and small foreign exchange markets see a very large impact of a billion dollars of intervention, as in Cape Verde, Guyana and Gambia. In other words, the estimates of ρ – though requiring restrictive assumptions – conform to priors suggested by finance theory. Table 3 shows the estimated values of ρt and the daily turnover in the spot and forwards currency market in or around the same years: Brazil in 1997, India in 2001 and Malaysia in 2002 (Unfortunately, foreign exchange turnover data is not available for most of the country periods for which ρt can be estimated). Trading in the foreign exchange market takes place on an average of 20 days a month. In the case of India, for example, the turnover in the market in one month in 2001 was USD 3 billion a day or USD 60 billion per month. Our estimates of ρ suggest that a billion dollars of trade per month by the Indian central bank would have led to a change in the rupee-dollar rate of 1.55 percent in a month in 2001. The estimation of ρt in Section 2.2 gives us values for ρt for only 39 countryperiods. The missing ρt s therefore need to be predicted on the basis of the size of the currency market. Without data on turnover for most country-periods, we proxy it by the size of the economy, financial sector development and integration of the economy with the world economy. Figure 5 explores the validity of the proxies in the country-periods for which turnover data and estimates of ρ exist. We expect that as the economy grows bigger, there are more foreign exchange transactions – both the size and the number of transactions would increase. Thus the turnover in the foreign exchange market would be greater. This is seen in the positive relationship between GDP and the turnover in the foreign exchange market. We would also expect that as the size of GDP and the foreign exchange market turnover increase, the impact of a billion dollars of intervention will be lower. The figure shows a negative relationship between ρt and GDP. We exploit these relationships to set up a regression model to predict the missing ρt . For prediction of missing ρt , since data for the size of the market is not available for all countries and all years, we use the variables that predict foreign exchange market turnover. These include GDP, inflation and various measures of openness Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 15 Working paper No. 189 Figure 5 Relationship between GDP, size of the market and GDP and ρt 8 6 4 2 0 Log GDP (Billion USD) 10 We expect an inverse relationship between ρt and size of the foreign exchange market, or, as the size of the foreign exchange market increases, a billion dollars of intervention by the central bank has a smaller impact. These graphs show that at higher levels of GDP, turnover in the foreign exchange market is higher. Further, at higher levels of GDP, we see that ρt is smaller. 0 2 4 6 8 4 2 0 GDP (Billion USD) 6 Log Currency market turnover (Billion USD per day) 0 1 2 3 4 5 6 ρt Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 16 Working paper No. 189 Table 4 Model for predicting missing ρt This table displays the various specifications of macro-variables which have been used to model and predict ρ. We use Model 4 for predicting ρ values. Wherever values of trade intensity or FDI to GDP are missing, we use Model 1 with only GDP to predict values of ρ Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 5.55∗ 8.85∗ 9.00∗ 6.22∗ 6.58∗ 4.68∗ 6.72∗ 6.72∗ (0.17) (1.02) (1.11) (1.55) (1.68) (0.51) (1.71) (1.71) −0.93∗ −0.93∗ −0.89∗ −0.90∗ −0.88∗ −0.83∗ −0.83∗ GDP −0.89∗ (0.06) (0.05) (0.06) (0.06) (0.07) (0.06) (0.08) (0.08) −0.67∗ −0.36 −0.35 −0.55 −0.55 Trade to GDP −0.72∗ (0.22) (0.23) (0.27) (0.29) (0.32) (0.32) Inflation −0.21 −0.24 −0.20 (0.18) (0.20) (0.19) Net FDI to GDP −0.25∗ −0.26 −0.33∗ −0.30∗ −0.30∗ (0.11) (0.14) (0.11) (0.12) (0.12) 0.47 LMFn (0.42) 0.24 LM F 2 n (0.21) N 46 44 37 38 33 35 31 31 R2 0.85 0.88 0.88 0.88 0.88 0.88 0.88 0.88 0.85 0.87 0.87 0.87 0.86 0.86 0.86 0.86 adj. R2 Resid. sd 0.80 0.74 0.75 0.75 0.76 0.75 0.76 0.76 LMFn is the Lane-Milesi-Ferreti index after subtracting official reserves Standard errors in parentheses ∗ indicates significance at p < 0.05 Intercept of the economy such as the trade to GDP ratio, foreign direct investment and assets and liabilities of the country measured by the Lane and Milesi-Ferretti (2007) measure. Table 4 shows various models for predicting ρt . We use model 4 as our base model as the adjusted R-squared does not increase as we add/remove variables in subsequent models. For countries for which financial sector data is not available, missing values of ρt s are predicted using only GDP data. Using this method we predict annual values of ρt for 172 countries for the years 1995 to 2011. 2.4. EMP estimates We now have an annual multi-country dataset of the conversion factor ρt required for measuring emp. For a monthly emp dataset we assume that the values of the conversion factor remains constant over each year: while financial market liquidity fluctuates from day to day, secular changes take place over multi-year time horizons reflecting GDP, internationalisation of the economy, and financial sector development. Values of ρ for the latest years for which they could not be predicted due to unavailability of data are assumed to remain unchanged for the last observed year. These ρ estimates are used to compute monthly emp for all countries in the database (excluding eurozone countries) for the period January 1995 to April 2015. Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 17 Working paper No. 189 As an example, Figure 6 juxtaposes our proposed emp measure for China with the emp index. In the pre-crisis years of the 2000s, China either witnessed reserve accumulation or the currency appreciated. Consequently, the direction of emp should be only one way. This is seen in our proposed measure, where values less than zero represent a pressure to appreciate. Our proposed measure captures the pressure on the renminbi to appreciate through the 2000s. In contrast, an attempt to calculate the conventional emp index for China gives rise to values of infinity for 1999-2002 as the Chinese renminbi was maintained at a fixed peg of 8.71 per dollar with no variation during this period. The fixed exchange rate with no variation makes the first term in calculation of emp index (1/σ∆e = ∞) take the value of infinity. Additionally, the uni-directional pressure on the renminbi to appreciate is not evident in the pre-crisis years of 2000’s in the emp index. Figure 7 plots our proposed emp measure calculated for China, India, Brazil and Egypt for the same period as Figure 2, which plots the emp Index. India and Brazil witnessed pressure to appreciate for most of the 2000s, with the direction of pressure changing after the 2008 global crisis. Our proposed measure shows this contrasting magnitude and direction of pressures faced by the four countries in 2000’s, while the emp index in Figure 2 seems to indicate that the experience of the four countries has been indistinguishable. Figure 8 shows the emp for Egypt. We see a high pressure on the Egyptian Pound to depreciate with the onset of the Arab Spring. We observe that in the pre-crisis years of 2000’s, the Egyptian pound witnessed a sustained pressure to appreciate. 3. Questions on validity of assumptions and estimates We now examine the various threats to validity. A number of questions may be raised about the data used in the prediction of emp. While the change in exchange rates (∆et ) is directly observed, other variables such as conversion factor ρt and intervention It have been estimated. For the conversion factor ρt , we first estimated ρt for a small set of countries and then predicted ρt for all countries, across time periods, using their determinants, such as GDP. The variable for Central Bank intervention It has been estimated by change in reserves. In this section, we address the following questions regarding the validity of our assumptions and accuracy of our estimates: 1. ρt estimation: How sensitive are ρt estimates to macroeconomic shocks? 2. ρt prediction: How good are the predicted ρt s? 3. Intervention It estimation: How close are the emp measures in case of countries which publish monthly intervention data? 3.1. Is the estimation of ρt ’s sensitive to macroeconomic shocks? Though we have dropped country periods for crisis years and freely falling years, it is possible that countries may be moving from fixed to floating because of Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 18 Working paper No. 189 Figure 6 Comparison between proposed emp measure and emp Index for China This panel shows a comparison between our proposed emp measure and emp Index calculated for China for the time period 2000-2010. Our proposed emp measure suggests that the renminbi has faced increasing pressure to appreciate in pre-crisis years of 2000’s. Apart from 4 months when the renminbi depreciated, the renminbi either appreciated or was prevented from appreciating by intervention in foreign exchange markets. The emp index takes the value of infinity for the period 1999-2002 as σ∆e = 0, while our proposed measure works sensibly all through. 10 5 0 −5 −10 Per cent change in exchange rate Proposed measure 2000 2002 2004 2006 2008 2010 10 EMP Index 0 −10 −5 Index 5 EMP Index takes value of infinity 2000 2002 2004 Accessed at http://www.nipfp.org.in/publications/working-papers/1779 2006 2008 2010 Page 19 Working paper No. 189 Figure 7 Proposed emp measure for selected countries 20 10 0 −10 2005 2006 2007 2008 20 10 2004 Per cent change in exchange rate Brazil 2004 0 2009 2009 Accessed at http://www.nipfp.org.in/publications/working-papers/1779 2005 2006 2007 2008 2009 2008 2009 Egypt 20 2008 10 2007 0 2006 −10 2005 India −10 20 10 0 2004 Per cent change in exchange rate Per cent change in exchange rate China −10 Per cent change in exchange rate This panel shows our proposed emp measure for China, India, Brazil and Egypt calculated for the same time period as in figure 1. The figures show a consistent appreciation pressure on the currencies prior to the GFC, consistent with the direction of capital flows during this period 2004 2005 2006 2007 Page 20 Working paper No. 189 Figure 8 Proposed emp measure for Egypt This panel shows our proposed emp measure calculated for Egypt for the time period between 2004-2013. In the pre-crisis years of the 2000’s, we see a consistent pressure on the Egyptian pound to appreciate. After Arab spring, we observe a high pressure on the Egyptian pound to depreciate. 20 10 0 −10 Per cent change in exchange rate Arab Spring 2004 2006 2008 Accessed at http://www.nipfp.org.in/publications/working-papers/1779 2010 2012 Page 21 Working paper No. 189 macroeconomic shocks. If so, this would imply that the currency volatility in the two sets of floating periods, one that precedes and one that follows a fixed regime, would be different. We test whether such a difference exists using the Welch two-sample t-test and the two-sample Kolmogorov-Smirnov tests but we find no significant difference in either the means or the distributions. The values for the tests comparing the means and the distribution of the volatility of the exchange rate during the floating period are as follows: The t-test gave us a t-value of -1.66 with a p-value of 0.1 with 40 degrees of freedom. The value of the KolmogorovSmirnov statistic was 0.26 with a p-value of 0.35. Therefore, we consider both fixed to float and float to fixed episodes in estimating ρt . In the calculations for ρt we assumed that the two adjacent periods under consideration had similar macroeconomic volatility. If this assumption is true, then we would expect that macroeconomic shocks should not explain ρt . If we regress the calculated ρt s on various measures of macroeconomic shocks, the coefficients of these shocks should not be significant. Table 5 shows that ρt is not sensitive to variables such as inflation and the current account. We control for GDP, trade integration and capital flows which influence the size of the market and determine ρt . None of the coefficients are significantly different from zero. This suggests that the 39 regime changes that were used for estimation of ρt were in periods that were not periods of crisis or macroeconomic instability. 3.2. How good are the predicted ρt ’s ? To examine the goodness of our prediction strategy, we now compare the predicted measures of ρt using the above model to those originally estimated using the volatilities of the exchange rate and intervention. The comparison can be made only for 38 country-years for which ρt could be estimated. Table 4 shows that the predicted ρ values are in the same order of magnitude as the estimates. Figure 9 shows that the predicted values of ρ against available estimates are quite close. Figure 10 shows the correlation between the estimated and the model predicted ρt . The two are close to being on a 45 degree line. We test the stability of the predicted values of ρ by using prediction intervals to get a sense of the probability space of the true ρ parameter. We estimate ±σ prediction intervals for ρ and calculate upper and lower bounds for the ρ estimate (Figure 11). Predicted ρ values are being used to estimate emp. This necessarily introduces statistical imprecision in the resulting emp values. We setup a simulation where many draws from the distribution of ρ are utilised to obtain corresponding draws from the distribution of emp. Figure 12 superposes the 68% confidence interval with the emp estimate for China. This shows that while the estimate for each month has a wide confidence interval, the overall picture is still useful. In the public release of the dataset, we also release these confidence intervals for ρ and for emp. Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 22 Working paper No. 189 Table 5 The assumption of macroeconomic stability We test the sensitivity of the estimated ρt to various measures of macroeconomic shocks across different specifications of a model explaining the ρs. The coefficients for macroshocks are not significant and this suggests that assumption of macro-stability across our set of corresponding currency regimes holds. Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 5.00∗ 3.39∗ 3.26∗ 6.22∗ 6.58∗ 6.31∗ 5.95∗ (0.90) (0.32) (0.31) (1.55) (1.68) (1.92) (0.45) Inflation −0.84 −0.24 −0.23 (0.47) (0.20) (0.21) CA balance −0.00 (0.04) CAD to GDP −0.06∗ 0.01 0.01 (0.03) (0.02) (0.02) Trade Int −0.36 −0.35 −0.37 (0.27) (0.29) (0.32) GDP −0.89∗ −0.90∗ −0.92∗ −0.90∗ (0.06) (0.07) (0.07) (0.07) −0.26 −0.27 FDI to GDP −0.25∗ (0.11) (0.14) (0.14) N 39 41 41 38 33 36 36 R2 0.08 0.00 0.11 0.88 0.88 0.88 0.84 0.05 −0.03 0.09 0.87 0.86 0.86 0.83 adj. R2 Resid. sd 2.03 2.04 1.92 0.75 0.76 0.75 0.83 Standard errors in parentheses ∗ indicates significance at p < 0.05 Intercept Table 6 Comparing selected estimated ρt and predicted ρt Table compares estimated values of ρ with the predicted values of ρ. The predicted values appear comparable and in line with the ρ estimates Country Year Estimated ρt Predicted ρt India 2001 1.55 1.90 Malaysia 2001 5.35 4.64 Turkey 2002 4.42 3.64 Brazil 1997 1.97 1.14 Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Vietnam 2001 6.42 9.13 Kenya 2002 105.60 54.97 Sri Lanka 2001 28.20 27.09 Page 23 Working paper No. 189 Figure 9 Selected countries: Predicted and estimated values of ρt 5.0 Figure compares estimated values of ρ with the model predicted values of ρ for India, Brazil and Turkey. The model predicted values are comparable and in line with the ρ estimates 2.0 1.97 1.0 1.55 0.5 ρ units Predicted ρ: India Predicted ρ: Brazil Predicted ρ: Turkey Estimated ρ: India Estimated ρ: Brazil Estimated ρ: Turkey 4.42 1995 2000 2005 2010 2015 3.3. How robust is the measure of foreign exchange intervention? Intervention is not reported by most central banks. Consequently the literature uses the change in reserves as a proxy for intervention.6 But changes in reserves may also happen due to interest payments, or due to revaluation effects. It is not possible to accurately adjust for these without knowing the exact composition of reserves or the timing of interest payments. Further, intervention may be done through swaps, credit lines or intervention in derivatives markets which may not immediately affect reserve levels, but this data is usually not publicly available. In figure 13, we show that when actual intervention data is used for the countries for which central banks release data, the estimates of emp do not differ significantly from the measures obtained by using the change in reserves.We find central bank intervention time series for 6 countries; India, Brazil, South Korea, Mexico, Russia and Peru. We estimate emp for these countries using this intervention data and by using the change in reserves data and compare the two measures. The two measures appear similar. This also corroborates recent work by Suardi and Chang (2012) who suggest that changes in reserves are a reliable proxy for central bank intervention. 6 See Pentecost et al. (2001), Sachs et al. (1996), Kaminsky et al. (1998) and Eichengreen et al. (1996) Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 24 Working paper No. 189 Figure 10 Scatter plot of model predicted ρt versus estimated ρt 20 50 1 2 5 Estimated ρ 200 This figure shows all estimated values of ρ with the corresponding model predicted values of ρ. The predicted values are correlated with the ρ estimates and are close to the 45 degree line 1 5 10 50 100 500 Predicted ρ Figure 11 Predicted ρt with ±σ prediction intervals This figure shows predicted values of ρ with ±σ prediction intervals. The fitted values of ρ are close to the estimated values of ρ all cases and lie within the 68% prediction interval. 0.20 1.00 1.55 0.05 ρ units 5.00 China 68% PI China India 68% PI India Estimated ρ: India 1995 2000 2005 Accessed at http://www.nipfp.org.in/publications/working-papers/1779 2010 2015 Page 25 Working paper No. 189 Figure 12 emp index for China with confidence interval Point estimate 68% confidence interval 10 ● ● ● ● ● ● ● ● ●● 0 ● ● ● ● ● ●● ● ●●● ● −10 % change in exchange rate 20 This figure plots the emp index for China. The dots represent the point estimate and the lines represent the 68% confidence interval ● ● ●● ● ● ● ● ● ● ●●●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −20 ●● 2004 2006 2008 2010 2012 4. Reproducible research A database with monthly data for the proposed emp measure, along with the computer programs used in this research, have been placed on the web7 . The authors hope this will enable replication and downstream research. The data is available for 139 countries and spans from 1995 to 2012 for most countries (due to limited data availability for some countries). The authors propose to update this database four times a year, and thus make it a useful resource for researchers. 5. Conclusion Previous emp measures were employed largely to predict crises. They gave misleading results in more tranquil periods, and could not be used for cross-country analysis. In this paper we develop a new emp measure that can be used in normal times, and permits panel data analysis. Since exchange rate changes and intervention are in different units, the paper focuses on creating a conversion factor that allows both to be measured in terms of exchange rate changes, i.e. the change that occurred and the change that was prevented by intervention. 7 http://macrofinance.nipfp.org.in/releases/exchange market pressure.html Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 26 Working paper No. 189 Figure 13 Intervention data for emp calculations: Actual versus change in reserves India −10 0 10 20 EMP with change in reserves EMP with reported intervention −20 Per cent change in exchange rate 30 This figure compares estimates of emp calculated with reported intervention and change in reserves for India, Brazil and Korea. This suggests that change in reserves are a good proxy for intervention data 2004 2006 2008 2010 2012 2014 2010 2012 2014 2010 2012 2014 Brazil −10 0 10 20 EMP with change in reserves EMP with reported intervention −20 Per cent change in exchange rate 30 2002 2004 2006 2008 South Korea −10 0 10 20 EMP with change in reserves EMP with reported intervention −20 Per cent change in exchange rate 30 2002 2002 2004 2006 2008 Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 27 Working paper No. 189 Such a counterfactual can, of course, not be measured accurately. We provide an estimate of the exchange rate change that was prevented, based on a series of restrictive assumptions, most notably that intervention has systematic and durable effects on exchange rate levels, which are related to the size of the market. The dataset has been released in the public domain and opens up many new academic and policy research possibilities. 6. Appendix 6.1. Defining exchange rate regimes Estimates of the conversion factor depends upon the transitions from pegs to float and float to pegs. We use Zeileis et al. (2010) (ZSP) to identify these periods. In this Appendix, we show that the mapping used from the much more familiar Reinhart and Rogoff (2004) (RR) classification to the R2 calculated for different country periods by ZSP. Table 7 Comparing RR and ZSP across both datasets We compare the R2 calculated for different country periods using ZSP methodology for 137 countries with the RR coarse classification. We compare the RR score with the ZSP R2 of the Frenkel-Wei regression to ascertain the R2 thresholds between different de-facto currency regimes RR Score 1 2 3 4 5 6 RR Classification Peg Crawling Pegs Managed Floats Free Floats Freely Falling Multiple Arrangements Average ZSP R2 Max R2 Min R2 0.85 1 0.06 0.81 1 0.16 0.61 1 0.08 0.54 1 0.03 0.57 1 0.03 0.88 1 0.44 Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 28 Working paper No. 189 Table 8 Reinhart and Rogoff (2004) monthly-coarse classification This table describes the Reinhart-Rogoff monthly-coarse currency classification. Code 1 1 1 1 2 2 2 2 3 3 3 3 4 5 6 Description No separate legal tender Pre announced peg or currency board arrangement Pre announced horizontal band that is narrower than or equal to +/-2% De facto peg Pre announced crawling peg Pre announced crawling band that is narrower than or equal to +/-2% De factor crawling peg De facto crawling band that is narrower than or equal to +/-2% Pre announced crawling band that is wider than or equal to +/-2% De facto crawling band that is narrower than or equal to +/-5% Moving band that is narrower than or equal to +/-2% (i.e., allows for both appreciation and depreciation over time) Managed floating Freely floating Freely falling Dual market in which parallel market data is missing. Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 29 Working paper No. 189 Table 9 Comparing RR and ZSP for float periods used in the paper The table shows float periods detected by the Zeileis et al. (2010) (ZSP) methodology and compares it with the Reinhart and Rogoff (2004) (RR) de facto coarse currency classification. Majority of the country periods which are detected as floats by the ZSP methodology are categorized as crawling pegs or managed floats by RR database Country Angola Bangladesh Brazil Cape Verde Ethiopia Guinea Guyana Guyana India Kenya Kazakhstan Laos Sri Lanka Mongolia Maldives Malaysia Tunisia Trinidad & Tobago Venezuela Antigua & Barbuda Angola Costa Rica Cape Verde Gambia Guyana Guyana Moldova Mauritius Malaysia Tunisia Ukraine Central African Republic Float Period Nov 2006 to May 2007 Dec 2005 to Jan 2007 Jun 1994 to Jul 1995 Mar 1999 to Sep 2001 Sep 2002 to May 2007 Aug 1998 to Sep 1999 Oct 1998 to Jul 1999 Jun 2005 to Dec 2005 Aug 1997 to Aug 1998 Apr 1997 to Jul 2001 Mar 2006 to Sep 2007 Jun 2001 to Nov 2001 Jun 2000 to Jun 2001 Sep 1998 to Mar 2001 May 2005 to Apr 2006 Aug 1997 to Aug 1998 Sep 1990 to Sep 1991 Sep 1996 to Oct 1997 Feb 2002 to Sep 2003 Feb 1996 to Aug 2002 Nov 2006 to May 2007 Jan 1997 to Jul 1997 Jul 2004 to Dec 2007 Dec 1998 to Nov 2003 Jun 2005 to Dec 2005 Jan 2007 to Jul 2007 Nov 2000 to May 2001 Dec 2002 to Apr 2004 Dec 1993 to Jul 1994 Aug 1992 to Jan 1994 Apr 2003 to Feb 2004 May 2002 to Jan 2004 Accessed at http://www.nipfp.org.in/publications/working-papers/1779 ZSP R2 0.55 0.62 0.51 0.31 0.65 0.55 0.48 0.49 0.5 0.54 0.58 0.44 0.48 0.45 0.46 0.21 0.47 0.59 0.34 0.63 0.55 0.41 0.44 0.5 0.49 0.47 0.56 0.62 0.44 0.61 0.59 0.5 RR Classification 1 2 2 2 2 2 2 2 2 2 2 6 3 1 1 4 2 2 4 1 1 2 2 2 2 2 2 2 2 2 1 1 Page 30 Working paper No. 189 References BIS, 1993. 63rd Annual Report. Technical Report. Bank for International Settlements. 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, 383–402. Eichengreen, B., Rose, A., Wyplosz, C., 1996. Contagious currency crises. Technical Report. National Bureau of Economic Research. Evans, M.D.D., Lyons, R.K., 2006. Understanding order flow. International Journal of Finance & Economics 11, 3–23. URL: http://ideas.repec.org/a/ijf/ ijfiec/v11y2006i1p3-23.html. Frankel, J.A., Wei, S.J., 1994. Yen bloc or dollar bloc? exchange rate policies of the east asian economies, in: Macroeconomic Linkage: Savings, Exchange Rates, and Capital Flows, NBER-EASE Volume 3. National Bureau of Economic Research, Inc. NBER Chapters, pp. 295–333. URL: http://ideas.repec.org/ h/nbr/nberch/8537.html. Girton, L., Roper, D., 1977. A monetary model of exchange market pressure applied to the postwar canadian experience. The American Economic Review 67, 537–548. Guimares, R.P.P., Karacadag, C., 2004. The empirics of foreign exchange intervention in emerging markets: the cases of Mexico and Turkey. volume 4. International Monetary Fund. IMF, 2007. Managing large capital inflows. Technical Report. International Monetary Fund. Kaminsky, G., Lizondo, S., Reinhart, C., 1998. Leading indicators of currency crises. Staff Papers-International Monetary Fund , 1–48. Klaassen, F., 2011. Identifying the Weights in Exchange Market Pressure. Tinbergen Institute Discussion Papers 11-030/2. Tinbergen Institute. URL: http://ideas.repec.org/p/dgr/uvatin/20110030.html. Klaassen, F., Jager, H., 2011. Definition-consistent measurement of exchange market pressure. Journal of International Money and Finance 30, 74–95. Lahura, E., Vega, M., 2013. Asymmetric effects of FOREX intervention using intraday data: evidence from Peru. BIS Working Papers 430. Bank for International Settlements. URL: https://ideas.repec.org/p/bis/biswps/430.html. Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 31 Working paper No. 189 Lane, P., Milesi-Ferretti, G., 2007. The external wealth of nations mark II: Revised and extended estimates of foreign assets and liabilities, 1970-2004. Journal of International Economics 73, 223–250. URL: http://www.sciencedirect.com/ science/article/pii/S0022199607000591. Menkhoff, L., 2013. Foreign exchange intervention in emerging markets: A survey of empirical studies. The World Economy . Pentecost, E., Van Hooydonk, C., Van Poeck, A., 2001. Measuring and estimating exchange market pressure in the eu. Journal of International Money and Finance 20, 401–418. Pontines, V., Siregar, R., 2008. Fundamental pitfalls of exchange market pressurebased approaches to identification of currency crises. International Review of Economics & Finance 17, 345–365. Reinhart, C.M., Rogoff, K.S., 2004. The modern history of exchange rate arrangements: A reinterpretation. The Quarterly Journal of Economics 119, 1–48. URL: http://ideas.repec.org/a/tpr/qjecon/v119y2004i1p1-48.html. Sachs, J., Tornell, A., Velasco, A., 1996. Financial crises in emerging markets: the lessons from 1995. Technical Report. National Bureau of Economic Research. Scalia, A., 2008. Is foreign exchange intervention effective? some microanalytical evidence from the czech republic. Journal of International Money and Finance 27, 529–546. URL: http://ideas.repec.org/a/eee/jimfin/ v27y2008i4p529-546.html. Suardi, S., Chang, Y., 2012. Are changes in foreign exchange reserves a good proxy for official intervention? Journal of International Financial Markets, Institutions and Money . Tapia, M., Tokman, A., 2004. Effects of Foreign Exchange Intervention Under Public Information: the Chilean Case. Working Papers Central Bank of Chile 255. Central Bank of Chile. URL: http://ideas.repec.org/p/chb/bcchwp/ 255.html. Weymark, D., 1995. Estimating exchange market pressure and the degree of exchange market intervention for canada. Journal of International Economics 39, 273–295. Zeileis, A., Shah, A., Patnaik, I., 2010. Testing, monitoring, and dating structural changes in exchange rate regimes. Computational Statistics & Data Analysis 54, 1696–1706. Accessed at http://www.nipfp.org.in/publications/working-papers/1779 Page 32 MORE BY THE AUTHORS • Pandey R. and Patnaik, I, (2016). Legislative Strategy for Setting up an Independent Debt Management Agency, WP No. 178 (October) • Pandey, R. , Patnaik, I., and Shah, A., (2016). Dating Bussiness Cycles in India, WP No. 175 (September) Ila Patnaik, is Professor, NIPFP Email: [email protected] Joshua Felman, IMF Email: [email protected] com Ajay Shah, is Professor, NIPFP Email: [email protected] MORE IN THE SERIES • Mundle, S. (2017). Employment, Education and the State, WP No. 188 (February) • Mundle, S. (2017). Beyond Catch Up: Some Speculations About the Next Twenty Five, WP No. 187 (January) • Chhibber, A. and Gupta S., (2017). Public Sector Undertakings – Bharat’s Other Ratnas, WP No. 186 (January) National Institute of Public Finance and Policy, 18/2, Satsang Vihar Marg, Special Institutional Area (Near JNU), New Delhi 110067 Tel. No. 26569303, 26569780, 26569784 Fax: 91-11-26852548 www.nipfp.org.in