EARLY WARNING SYSTEM AND CURRENCY
VOLATILITY MANAGEMENT IN EMERGING MARKET
Natasia Engeline S1
Salomo Posmauli Matondang2
Abstract
This paper adopts theoretical models from Candelon, Dumitrescu, and Hurlin and empirical model from Commerzbank to devise a set of indicators that can serve as an early warning system (EWS) on exchange rate. In light of the appreciation of emerging countries’ currencies during the Fed quantitative easing period, it is important to understand on how The Fed normalization would put pressure on managing volatility for central banks, especially for countries with large trade and fiscal deficit such as Indonesia. All in all, using both EWS models, central banks could discern potential exchange rate depreciation for intervention purpose.
Keywords: Dynamic Logit Model, Foreign Exchange, Early Warning System, Emerging Countries, Foreign Exchange Intervention.
JEL Classification: C32, E58, F31, F37.
1Financial Analyst, Reserve Management Department of Bank Indonesia (natasia@bi.go.id)
2Financial Analyst, Reserve Management Department of Bank Indonesia (salomo@bi.go.id)
130Buletin Ekonomi Moneter dan Perbankan, Volume 19, Nomor 2, Oktober 2016
I. INTRODUCTION
Research on emerging countries’ exchange rate dynamics has been the subject of interest by academia, market participants, and financial regulators; considering their high volatility. Central banks especially those in emerging market economies are taking special interest in exchange rate dynamics due to their role in stabilizing exchange rate movement through intervention. Gracia et al. (2011) argued that the role of central bank intervention in exchange rate movement is desirable, especially in vulnerable emerging economies. Alder and Tovar (2011), Basu and Varoudakis (2013), and Neely (2008), identified several motives of the central bank’s intervention such as moderate exchange rate volatility, reducing exchange rate misalignment, accumulate reserves, and supply foreign exchange to the market.
More recently, following the unconventional policies of major advanced economies from 2008, emerging countries’ currencies have experienced appreciation due to massive capital inflow. Looking forward, it is important to understand on how The Federal Reserve normalization would put pressure for central banks on managing volatility, especially those with large trade and fiscal deficit such as Indonesia. In retrospect, initial reports that The Federal Reserve might begin “tapering” its quantitative easing on May 2013, caused a rush to exit from emerging countries including Indonesia, with exchange rate declines of as much as 20% in the following four months.
Considering the importance of intervention in managing emerging currencies, central banks in emerging market economies should devise a set of indicators that can serve as an early warning system (EWS); which could identify an impending depreciation before it occurs. EWS could help central banks implement optimal policies including the strategies of intervention to prevent or smooth the impact of currency depreciation.
Kaminsky, Lizondo, and Reinhart (1998) pioneered a comprehensive survey regarding EWS by proposing several case studies of devaluation episodes using structural model of balance of payment crises, signaling model for currency crises, as well as empirical study using macroeconomic and financial data for emerging countries. Berg and Patillo (1999) proposed a static panel probit model as an alternative to the signaling approach. Bussiere and Fratzcher (2006) proposed a multinominal logit EWS that consider the crisis as a ternary variable instead of binary.
Unfortunately, in previous studies, EWS have remained silent at the recent financial crisis. The difficulty to detect potential currency depreciation lies in the specificity of EWS that aimed at accurately detecting the occurrence of a currency depreciation which is translated into a binary variable that takes the value of one when depreciation occurs and the value of zero otherwise. In this context, it is not possible to directly implement the method proposed in times series econometrics such as vector autoregression. Furthermore, most previous EWS are static and assume that the depreciation probability depends only on a set of macroeconomic variables.
Early Warning System and Currency Volatility Management in Emerging Market 131
Candelon, Dumitrescu, and Hurlin (2010) proposed a new generation of EWS which reconcile the limited dependent property of the depreciation variable and the dynamic dimension of this phenomenon. In particular, Candelon et al. (2010) considered not only the exogenous source of depreciation persistence from macroeconomic data, but also endogenous persistence of depreciation which are lagged binary depreciation variable and past index associated to the probability of being in depreciation period. Thus, the EWS relies an autoregressive (AR) model, where the endogenous variable summarizes all the past information of the system. Given all these different specifications, an exact maximum likelihood estimation by Kauppi and Saikonnen (2008) is used to estimate the models.1
In contrast from academic EWS model, Commerzbank (2013) proposed a simple currency depreciation index that requires a shorter forecast period and does not require regular recalibration. Commerzbank used several macroeconomics indicators such as current account and industrial production as well as market indicators such as real effective exchange rate and equity market performance that translated into risk measures with equal weighting.
Perhaps the most interesting feature of our research is on how we adopt both models from Candelon et al. (2010) and Commerzbank (2013) to give a better understanding toward potential currency depreciation. All in all, using both EWS model, central banks could discern potential exchange rate depreciation for intervention purpose.
This paper is structured as follows. Section 2 describes the structure of early warning signal (EWS) by Candelon et al. (2010) and Commerzbank as well as several assumptions for the EWS index. Section 3 describes the data and construction of the EWS index. Section 4 reports the forecast evaluation and intervention strategies, while section 5 concludes.
II. THEORY
The first model is a dynamic EWS based on Candelon et al. (2010) that exploits the persistence property of the currency depreciation captured by lagged endogenous indicators. The second model is based on Commerzbank that use macroeconomics and market indicators to construct a depreciation warning index.
2.1. Dating Currency Depreciation
Before elaborating further into the EWS model, we define currency depreciation as large market movement adjusted for interest rate differentials rather that looking at composite indices of exchange rate pressure as elaborated by Kumar et al. (2002). Thus, if et is the exchange rate
1 Detail explanation on constrained maximum likelihood is available on appendix.
132Buletin Ekonomi Moneter dan Perbankan, Volume 19, Nomor 2, Oktober 2016
where γ1 is a
Furthermore, second definition of currency depreciation or known as depreciation crashes could also be defined if
Where γ2 is a
2.2. Specification and Estimation of Dynamic EWS Model
First, consider the as the currency depreciation binary variable for country n, taking the value of 1 during
depreciation periods and 0 otherwise and |
as the matrix of explanatory variables, |
i.e., macroeconomic indicators. |
|
The
Early Warning System and Currency Volatility Management in Emerging Market 133
where
The main advantage of the general framework above is that it allows to estimate and to compare different alternative specifications taken the form as follows:
•Pure static model in which the occurrence of currency depreciation is explained only by
exogenous macroeconomic variables
•Dynamic model in which the occurrence of currency depreciation is explained by exogenous
macroeconomic variables and lagged value of the binary dependent variable
•Dynamic model in which the occurrence of currency depreciation is explained by exogenous
macroeconomic variables and lagged index
•Finally, the most complex dynamic model, including both the lagged dependent variable
Furthermore, since the last two models have δ as an autoregressive parameter, it has to satisfy the usual stationarity condition. Otherwise, the depreciation becomes perpetual, which is counterintuitive. In order to overcome this problem, a constrained maximum likelihood estimation is implemented which general form of the
134Buletin Ekonomi Moneter dan Perbankan, Volume 19, Nomor 2, Oktober 2016
where θ is the vector parameters. Given the
2.3. Specification and Estimation of Commerzbank Model
Commerzbank (2013) developed a simple and intuitive EWS model, using both macroeconomic indicator and market indicator, that requires shorter forecast period, does not require regular recalibration, and makes clear contribution of individual inputs to the overall risk signal.
Macroeconomics indicators that are being used to construct the index are as follows
•Current account: This gives an indication of the degree to which a country relies on foreign funding. Higher current account surplus may translate into lower volatility in the currency.
•Money supply: Excessive money creation may lead to higher inflation and consequently a weaker currency.
•Inflation: Excessive inflation will typically lead to depreciation of the currency.
•Industrial production: Falling industrial production may signal that the economy is weakening. Lower interest rate and/or a weaker currency may be required to stimulate a recovery thus triggering currency depreciation.
•Trade data (exports): A sharp fall in exports will lessen demand for the currency. A weaker currency may in any case be necessary to increase the competitiveness of the export market.
•Short term debt: High level of short term debt increases the risk of a funding crisis should debt become difficult to roll over.
•
•Domestic credit: Very high level or credit to the domestic private sector may indicate excesses in the banking system.
•Economic surprises: Worse than expected economic data may result from deterioration in the economy that could lead to the withdrawal of capital from local assets and consequent weakening of the currency.
Market indicators that are being used to construct the index are as follows
•Real effective exchange rate: The
Early Warning System and Currency Volatility Management in Emerging Market 135
•FX implied volatility: The level of FX implied volatility acts as a proxy for option prices and hence the approximate cost of hedging. An increased level of hedging activity may be indicative of concern over a weakening of the currency.
•Equity market performance: Weaker asset markets can lead to withdrawal of foreign capital. If investors repatriate the realized funds there will be selling pressure on the local currency.
•Global risk sentiment: The global risk environment can influence currency markets through the home bias effect – investors in developed markets are more likely to withdraw funds from emerging markets when risk is perceived as being high.
Commerzbank model use simple steps to generate a warning index as follows
•Rank each data point with respect to its own history.
•Convert percentile rankings into risk measures. Where a lower value is more likely to cause for concern, i.e., industrial production, the risk level is given by one hundred minus the percentile ranking, otherwise the risk measure is simply given by the rank.
•Risk rankings for all macroeconomic and market indicators for an individual country are simply averaged to generate an overall risk rating on a scale from 0 to 100 for the currency in question.
•Risk index is calculated using equal weighting
2.4. Optimal
In order to compare the depreciation probabilities obtained from EWS model with the actual currency depreciation, we have to shift these probabilities to depreciation forecasts by defying an optimal threshold or
136Buletin Ekonomi Moneter dan Perbankan, Volume 19, Nomor 2, Oktober 2016
We address this
Tabel 1
True versus predicted occurrence of depreciation
True Value
|
|
Depreciation |
Calm |
Total |
|
|
|
|
|
|
Depreciation |
True Positive |
False Positive |
All predicted depreciation |
Predicted results |
Calm |
False Negative |
True Negative |
All predicted calm |
|
Total |
All true depreciation |
All calm |
T (sample size) |
|
|
|
|
|
Sensitivity refers to the ability to correctly identify currency depreciation using a
Where specificity refers to the ability to correctly identify calm period using a
Optimal
III.METHODOLOGY 3.1. Dataset
For the dynamic EWS, the dataset covers Indonesia’s monthly data expressed in US dollar available from February 1999 to May 2015 and is extracted via Bloomberg. There are two macroeconomic variables used in the dynamic EWS model:
Early Warning System and Currency Volatility Management in Emerging Market 137
For the Commerzbank depreciation index, monthly data expressed in US dollar available from the period January 2004 to May 2015 is also extracted via Bloomberg. Taking concern about some limitation in the data availability, macroeconomic indicators used in the Commerzbank depreciation index are reduced to current account, money supply, inflation, industrial production, and
3.2. Model Evaluation and Robustness Test
In order to show the usefulness of the model, we implement the EWS evaluation by Candelon et al. (2011), especially to test their forecasting abilities
Accordingly, we consider both classic EWS evaluation measures such as the QPS criterion and newer one for the EWS literature, which take the
At the same time, AUC is a
forecasts :
Where Se(c) represents the sensitivity, i.e. the proportion of depreciation correctly identified by the EWS for a given
138Buletin Ekonomi Moneter dan Perbankan, Volume 19, Nomor 2, Oktober 2016
Next, the optimal
where J(c)=Se(c)+
IV. RESULT AND ANALYSIS
4.1. Estimation Results for Dynamic EWS
General form of dynamic EWS elaborated above has main advantage that it allows to estimate and compare different EWS specification. First, we estimate the three types of dynamic EWS as well as the benchmark which is the static EWS model under analysis in the
Second, we find the best goodness of fit from the four models by relying on Schwarz Information Criterion (SBC). SBC reveals that the
Tabel 2
SBC information criterion
Country |
Model |
1 |
Model |
2 |
Model |
3 |
Model |
4 |
|
SBC |
|
SBC |
|
SBC |
|
SBC |
|
Indonesia |
190,81 |
159,94 |
195,87 |
164,91 |
||||
|
|
|
|
|
|
|
|
|
Note: Model 1 is the static model (the benchmark), Model 2 to 4 are dynamic. Bold values correspond to the best model according to SBC.
Third, we analyze the signs of the estimated parameters for the Model 2. The result shows a negative coefficient of growth of international reserves, indicating a decline in the probability of currency depreciation is presumed with an increase in a country’s growth of international
Early Warning System and Currency Volatility Management in Emerging Market 139
reserves. Intuitively, an increase in growth of international reserves indicates currency non- vulnerability. For the growth of M2 to reserves coefficient, it is assumed that if the growth of the amount of money in circulation overruns the growth of international reserves, the currency is perceived as unstable and a speculative attack is foreseeable. Thus, a positive coefficient of the growth of M2 reserves is expected. Nonetheless, a negative coefficient that appears on growth of M2 to reserves might be due to the fact that the two macroeconomic variables capture mainly the information not filtered by the lagged binary variable. Most importantly, the coefficient of the lagged binary dependent variable is significant and has a positive sign. It means that the probability of being in a deprecation episode increases if a depreciation period prevailed in the previous period. This clearly indicates that depreciation’ persistence should be accounted for in order to improve accuracy of currency EWS.
Tabel 3
Estimation results
Country |
Indicator |
Model 1 |
Model 2 |
Model 3 |
Model 4 |
|
Intercept |
||||
|
|
(0.300) |
(0.316) |
(0.418) |
(0.478) |
|
Lagged binary variable |
|
2.598 |
|
2.743 |
|
|
|
(0.422) |
|
(0.517) |
Indonesia |
Growth of international reserves |
||||
|
|
(1.640) |
(1.494) |
(1.809) |
(2.037) |
|
Growth of M2 to reserves |
||||
|
|
(0.739) |
(1.317) |
(0.476) |
(1.393) |
|
Lagged index |
|
|
0.254 |
|
|
|
|
|
(0.361) |
(0.188) |
Note: Model 1 is the static model (the benchmark), Model 2 to 4 are dynamic. |
|
|
|
||
Robust standard errors are reported in the parentheses. |
|
|
|
|
Moreover, the signs are similar from one model to another, confirming the economic intuition that a higher growth of international reserves lowers the depreciation probability. On contrary, the M2 to reserves indicator is generally not significant.
4.2. Estimation Results for Commerzbank Index
Commerzbank model has main advantage that it is rather intuitive and doesn’t require regular calibration like the dynamic EWS. First we rank each data point from both macroeconomic and market indicator to its own history. Second, we convert percentile ranking into risk measures. Where a lower value is more likely to give cause for concern, i.e., industrial production, the risk level is given by one hundred minus the percentile ranking, otherwise the risk measure is simply given by the rank. Third, we average the risk ranking to generate an overall risk ranking.
140Buletin Ekonomi Moneter dan Perbankan, Volume 19, Nomor 2, Oktober 2016
Tabel 4
Risk ranking Commerzbank depreciation Index
|
|
Risk Level |
CA |
M2 |
Inflation |
IP |
NPL |
REER |
FXIV |
EMP |
SI |
Jul |
2014 |
50 |
90 |
10 |
39 |
81 |
84 |
26 |
55 |
34 |
30 |
Aug |
2014 |
45 |
90 |
12 |
22 |
33 |
79 |
28 |
51 |
64 |
30 |
Sep |
2014 |
45 |
86 |
18 |
40 |
18 |
54 |
29 |
70 |
69 |
17 |
Oct |
2014 |
49 |
86 |
21 |
49 |
38 |
75 |
28 |
53 |
72 |
17 |
Nov |
2014 |
50 |
86 |
23 |
68 |
50 |
57 |
42 |
51 |
59 |
17 |
Dec |
2014 |
49 |
86 |
18 |
100 |
38 |
8 |
47 |
69 |
57 |
18 |
Jan |
2015 |
57 |
86 |
37 |
78 |
44 |
97 |
48 |
63 |
59 |
5 |
Feb |
2015 |
59 |
86 |
56 |
67 |
73 |
82 |
42 |
73 |
44 |
5 |
Mar |
2015 |
52 |
84 |
60 |
69 |
26 |
49 |
39 |
75 |
59 |
6 |
Apr |
2015 |
57 |
84 |
43 |
75 |
35 |
81 |
39 |
56 |
94 |
6 |
May |
2015 |
52 |
84 |
31 |
81 |
36 |
80 |
35 |
60 |
50 |
15 |
Jun |
2015 |
57 |
83 |
31 |
83 |
36 |
80 |
35 |
56 |
89 |
15 |
|
|
|
|
|
|
|
|
|
|
|
|
Note: The heat map shows how IDR in terms of risk of depreciation. Darker colors indicate areas of greater. CA: current account,
M2: money supply, IP: industrial production, NPL:
volatility, EMP: equity market performance, SI: global risk sentiment.
The result shows an increasing risk ranking for current account, inflation,
|
|
|
|
|
|
|
|
|
|
x 104 |
80 |
|
|
|
|
|
|
|
|
|
1,4 |
|
|
|
Risk Level |
Optimal |
|
USDIDR |
||||
70 |
|
|
|
|
|
|
|
|
|
1,3 |
60 |
|
|
|
|
|
|
|
|
|
1,2 |
50 |
|
|
|
|
|
|
|
|
|
1,1 |
40 |
|
|
|
|
|
|
|
|
|
1 |
30 |
|
|
|
|
|
|
|
|
|
0,9 |
20 |
|
|
|
|
|
|
|
|
|
0,8 |
2005 |
2006 |
2007 |
2008 |
2009 |
2010 |
2011 |
2012 |
2013 |
2014 |
2015 |
Figure 1. Predicted probability of depreciation – in sample
(Commerzbank Index)
Early Warning System and Currency Volatility Management in Emerging Market 141
Historically, distribution of IDR depreciation index is centered on 45 to 60 and extreme values are seldom seen. Furthermore, the warning signal has on average come a few months earlier.
25
20
15
10
5 |
0 |
10 20 30 40 50
60 |
70 |
80 |
90 |
Figure 2.
Distribution of Commerzbank Index
4.3. Forecast Evaluation and Intervention Strategies
In this section, we go one step further and test the
4.3.1.
To check the within sample forecasting abilities of the static and dynamic
142Buletin Ekonomi Moneter dan Perbankan, Volume 19, Nomor 2, Oktober 2016
Tabel 5
Evaluation Criteria
|
|
|
Commerzbank Model |
|||
|
Static Logit |
Dynamic Logit |
||||
|
|
|
||||
|
QPS |
AUC |
QPS |
AUC |
QPS |
AUC |
Indonesia |
0,303 |
0,587 |
0,233 |
0,774 |
0,538 |
0,534 |
|
|
|
|
|
|
|
Note: The AUC criteria takes value between 0.5 and 1, 1 being the perfect model, while QPS ranges from 0 to 2, 0 being the perfect accuracy. Bold values correspond to the best model according to AUC and QPS.
To be more exact, we first compare the static and the dynamic logit models (SL vs. DL) and show that the dynamic
Tabel 6
Optimal
|
|
|
|
|
|
Commerzbank Model |
|||||
|
|
Static Logit |
|
Dynamic Logit |
|
||||||
|
|
|
|
|
|
|
|||||
|
Se |
|
Sp |
Se |
|
Sp |
Se |
Sp |
|||
Indonesia |
0,189 |
0,528 |
|
0,529 |
0,112 |
0,722 |
|
0,719 |
0,5 |
0,444 |
0,471 |
|
|
|
|
|
|
|
|
|
|
|
|
Note: We identify the optimal
The optimal
Our findings prompt the fact that there are gains from using a dynamic EWS specification. This includes the lagged binary depreciation indicator.
Early Warning System and Currency Volatility Management in Emerging Market 143
1 |
13000 |
|
0,9 |
12000 |
|
0,8 |
||
|
||
0,7 |
11000 |
|
0,6 |
10000 |
|
0,5 |
||
9000 |
||
0,4 |
||
|
||
0,3 |
8000 |
|
0,2 |
7000 |
|
0,1 |
||
|
||
0 |
6000 |
Jan 01 Sep 02 Mei 04 Jan 06 Sep 07 Mei 09 Jan 11 Sep 12 Mei 14
Observed Crisis |
Optimal |
|
EWS |
|
|
|
|
|
0,35 |
|
|
|
|
0,3 |
|
|
|
|
0,25 |
|
|
|
|
0,2 |
|
|
|
|
0,15 |
|
|
EWS |
|
0,1 |
|
|
|
|
|
|
|
Optimal |
0,05 |
|
|
|
USDIDR |
|
|
|
|
|
|
|
|
|
|
|
0 |
Jan 01 |
Sep 02 May 04 Jan 06 |
Sep 07 May 09 Jan 11 |
Sep 12 May 14 |
|
|
|
Time |
|
|
|
|
|
|
Figure 3. |
|
|
|
|
|
|
|
|
Predicted probability of depreciation – in sample (Static EWS) |
|
|
|
|||||
1 |
|
|
|
|
|
x 104 |
|
|
|
|
0,9 |
|
|
|
|
1,4 |
|
|
|
|
0,8 |
0,8 |
|
|
|
|
|
EWS |
Optimal |
|
USDIDR |
|
0,7 |
|
|
|
|
1,2 |
|
|
|
|
0,6 |
0,6 |
|
|
|
|
|
|
|
|
||
|
|
|
|
|
|
|
|
|
|
|
0,5 |
|
|
|
|
|
|
|
|
|
|
0,4 |
|
|
|
|
1 |
|
|
|
|
0,4 |
0,3 |
|
|
|
|
|
|
|
|
|
|
0,2 |
|
|
|
|
0,8 |
|
|
|
|
0,2 |
0,1 |
|
|
|
|
|
|
|
|
||
|
|
|
|
|
|
|
|
|
|
|
0 |
|
|
|
|
|
|
|
|
|
|
Jan 01 |
Sep 02 May 04 Jan 06 Sep 07 May 09 Jan 11 |
Sep 12 May 14 |
0,6 |
|
|
|
|
0 |
||
|
|
Time |
|
|
|
|
|
|
||
|
|
|
|
|
Jan 01 Sep 02 May 04 |
Jan 06 Sep 07 May 09 |
Jan 11 |
Sep 12 |
May 14 |
|
Observed Crisis |
Optimal |
EWS |
|
|
||||||
|
|
|
Time |
|
|
|
||||
|
|
|
|
Figure 4. |
|
|
|
|
|
|
|
|
Predicted probability of depreciation – in sample (Dynamic EWS) |
|
|
|
4.3.2.
We then check the
It results that when faced more than one month of currency depreciation period, the EWS forecasting probability is very high during depreciation periods. On the contrary, when faced only one period of depreciation, the model forecasting abilities are disappointing. It is also seen in the
144Buletin Ekonomi Moneter dan Perbankan, Volume 19, Nomor 2, Oktober 2016
x 104 |
|
|
1,4 |
|
1 |
EWS |
Optimal |
USDIDR |
1,2 |
|
|
|
|
0,5 |
1 |
|
|
0,8 |
|
0 |
Mar 11 |
Jan 12 |
Nop 12 |
Sep 13 |
Jul 14 |
|
|
Time |
|
|
Figure 5. Predicted probability of depreciation
–
It results that during January 2010 to April 2013, Commerzbank depreciation index have captured several false signal due to the fact that unconventional monetary policy by developed central banks have caused massive capital inflows seeking for higher yield regardless of macroeconomic condition. However, after The Fed announced that it might begin “tapering” its purchases of US treasuries, there was a rush for exits from Indonesia which was confirmed by the Commerzbank depreciation index. Since May 2013, IDR has experienced the hardest hit due to its unfavorable macroeconomic condition and capital outflow.
x 104 |
|
|
1,4 |
|
80 |
EWS |
Optimal |
USDIDR |
1,2 |
|
60 |
1 |
|
40 |
0,8 |
|
20 |
Oct 10 |
Aug 11 |
Jun 12 |
Apr 13 |
Feb 14 |
Dec 14 |
|
|
Time |
|
|
|
Figure 6. Predicted probability of depreciation
–
Early Warning System and Currency Volatility Management in Emerging Market 145
All in all, the dynamic EWS model have good forecasting abilities not only
4.4. Intervention Strategy
After estimating three different types of EWS for forecasting depreciation period, we then try to understand why and how central bank intervene in foreign exchange markets as elaborated by Chutasripanich and Yetman (2015). Before assessing intervention strategy, first let us identify the motives of central banks for intervention which has been elaborated by Adler et al. (2011), Basu et al. (2013) and Neely (2008). Generally, these motives can be grouped as follows:
•Leaning against the wind: Recent survey by BIS shows that the most common reason for emerging market central banks to intervene in foreign exchange market was to limit exchange rate volatility and smooth the trend path of the exchange rate (BIS (2013)). For example, Adler et al (2011) find that half of the central bank in their sample intervene to dampen exchange rate volatility.
•Reducing exchange rate misalignment: Central bank may wish to step into the foreign exchange market if they see that the current value appears to be either overvalue or undervalued. It is presumed that an exchange rate that is too strong could reduce a country’s competitiveness and too low can lead to unsustainable growth and inflation. However, this statement is probably understated, due to the fact that central bank knows that equilibrium value of the exchange rate is hard to measure and depreciating one currency to increase competitiveness might attract a “currency war” stigma.
•Managing or accumulating FX reserves: After Asian financial crisis, many central banks find the urge to accumulate reserves for defending their currencies during crisis. Some central banks officially announced that intervention would be conducted for the purpose of building reserves, for example Turkey, South Africa, Chile, and Mexico.
•Ensuring liquidity: Some central banks may conduct intervention to ensure adequate liquidity in order to counter disorderly markets and avoid financial stress. BIS survey shows that more than half of participating central banks intervened to provide liquidity in the foreign exchange market.
In order to assess the effectiveness of intervention strategy, Chutasripanich et al (2015) modeled a simple analytical framework for two most common intervention strategies: leaning against exchange rate misalignment and leaning against the wind. Their model assumed that the fundamental value of the exchange rate is the value at which the current account is equal to zero. However, active trading by
146Buletin Ekonomi Moneter dan Perbankan, Volume 19, Nomor 2, Oktober 2016
rate away from this value. For example, if speculators engage in the carry trade, their returns depend on the behavior of both the exchange rate and interest rates.
The model also assumed that foreign exchange interventions are sterilized so that central banks are exposed to exchange rate risk and carry costs when they intervene. The effectiveness of different intervention rules are then assess using across five criteria: stabilizing the exchange rate, reducing current account imbalances, discouraging speculation, minimizing reserves volatility and limiting intervention costs. Their finding could be summarized as follows:
•The actions of speculators can, under some circumstances, reduce the volatility of exchange rates but, even then, they tend to increase exchange rate misalignment.
•Intervention that reduces exchange rate volatility also reduces the risks of speculation, creating a feedback loop and potentially leading to high levels of speculation.
•Uncertainty about the fundamental value of the exchange rate results in foreign exchange intervention being less efficient.
•Leaning against the wind, which avoids the problem of having to estimate the fundamental value might reduce the volatility of the exchange rate but tends to increase exchange rate misalignment.
•The cost of the foreign exchange intervention will be especially large when exchange rate movements are driven by interest rate shocks since these drive a positive correlation between the stock of reserves and the carrying cost of those reserves.
•Relative to transparent intervention, adding element of opaqueness offers both cost and benefits. It tends to increase the volatility of exchange rate, current account balances and reserves, but reduce the size of speculative flows and the cost of carrying reserves.
Comparison of the performance of intervention strategies to the shock of fundamental value of exchange rate (labelled ε), to the shock of interest rate differential (labelled δ), as well as both to the shock of fundamental and interest rate differential across five criteria, shows that there are no one approach that dominates.
Tabel 7
Optimal
|
|
|
|
|||||
|
Dynamic Logit |
|
Commerzbank Model |
|||||
|
Se |
|
Sp |
|
Se |
Sp |
||
Indonesia |
0,127 |
0,538 |
|
0,818 |
0,45 |
|
1,00 |
0,208 |
|
|
|
|
|
|
|
|
|
Note: The values of the
Early Warning System and Currency Volatility Management in Emerging Market 147
Tabel 8
Comparison of intervention strategies performance (Chutasripanich, 2015)
|
Strategies |
|
|
|
|
e and d |
Leaning against the wind |
Leaning against exchange |
|
|
rate misalignment |
Transparent opaque Transparent opaque
Objectives |
Stabilize exchange rate |
|
x |
x |
x |
Reduce current account imbalances |
x |
x |
|
x |
|
Reduce speculation |
x |
x |
|
x |
|
Reduce reserve volatility |
x |
|
|||
Minimize cost |
x |
|
|
|
|
|
|
|
|
|
Strategies |
|
|
|
|
e |
Leaning against the wind |
Leaning against exchange |
|
|
rate misalignment |
Transparent opaque Transparent opaque
Objectives |
Stabilize exchange rate |
|
x |
x |
x |
Reduce current account imbalances |
x |
x |
|
x |
|
Reduce speculation |
x |
x |
|
x |
|
Reduce reserve volatility |
x |
|
|||
Minimize cost |
|
|
|
|
|
|
|
|
|
|
|
Strategies |
|
|
|
|
d |
Leaning against the wind |
Leaning against exchange |
|
|
rate misalignment |
Transparent opaque Transparent opaque
Objectives |
Stabilize exchange rate |
|
x |
|
x |
Reduce current account imbalances |
|
x |
|
x |
|
Reduce speculation |
x |
x |
x |
x |
|
Reduce reserve volatility |
x |
x |
|||
Minimize cost |
x |
|
x |
|
|
|
|
|
V. CONCLUSION
Considering the importance of intervention in managing emerging currencies, this paper provides two EWS models that could be used to discern potential exchange rate depreciation for intervention purpose. In addition, this paper also outline several intervention strategies and their effectiveness in order to prepare for The Fed normalization that would put pressure on managing volatility for central banks, especially those with large trade and fiscal deficit such as Indonesia.
148Buletin Ekonomi Moneter dan Perbankan, Volume 19, Nomor 2, Oktober 2016
Several conclusions can be drawn from using both the dynamic EWS and Commerzbank index as well as incorporating them into intervention strategies. First, we show that in the in- sample test, dynamic logit models (sensitivity and specificity 72,2% and 71,9%) outperform static one (sensitivity and specificity 52,8% and 52,9%) as well as Commerzbank depreciation index (sensitivity and specificity 44,4% and 46,2%) . Second, by combining both EWS, we could have a better predictive ability of potential currency depreciation. Since the dynamic EWS give better predictive ability both within
Looking ahead, continuing pressures on IDR is inevitable. Nonetheless, there is no doubt that using both EWS model, central bank could implement optimal policies including the strategies of intervention to prevent or smooth the impact of currency depreciation.
Early Warning System and Currency Volatility Management in Emerging Market 149
REFERENCES
Adler, G., and Tovar, C.E. (2011). Foreign Exchange Intervention: A Shield Against Appreciation Winds?. IMF Working Paper 11/165.
Bank for International Settlements. (2005). Foreign Exchange Market Intervention in Emerging Markets: Motives, Techniques and Implication. BIS Papers 24.
Bank for International Settlements. (2013). Market Volatility and Foreign Exchange Intervention in EMEs: What has Changed?. BIS Papers 73.
Basu, K., and Varoudakis, A. (2013). How to Move the Exchange Rate if You Must: The Diverse Practice of Foreign Exchange Intervention by Central Banks and a Proposal for Doing it Better. World Bank Policy Research Working Paper 6460.
Berg, A., and Pattillo, C. (1999). Predicting Currency Crises: The Indicators Approach and an Alternative. Journal of International Money and Finance, 18,
Bussiere, M., and Fratzscher, M. (2006). Towards a New Early System of Financial Crises. Journal of International Money and Finance, 25(6),
Candelon, B., Dumitrecu, E.I., Hurlin, C. (2009). How to evaluate an Early Warning System? Towards a Unified Statistical Framework for Assessing Financial Crises Forecasting Methods. Working Paper.
Candelon, B., Dumitrecu, E.I., Hurlin, C. (2010). Currency Crises Early Warning System: why they should be Dynamic. Working Paper.
Commerzbank. (2013). Emerging Market Currency Exposure: How to Hedge and Manage Exposures to High Yield, Volatile Currencies. Commerzbank.
Falcetti, E., Tudela, M. (2006). Modelling Currency Crises in Emerging Markets: A Dynamic Probit Model with Unobserved Heterogeneity and Autocorrelated Errors. Oxford Bulletin of Economics and Statistics, 68(4),
Frankel, J.A., Yetman, J. (1990). Chartists, Fundamentalist and Trading in The Foreign Exchange Market. American Economic Review, 80(2),
Fuertes, A.M., Kalotychou, E. (2007). Optimal Design of Early Warning Systems for Sovereign Debt Crises. International Journal of Forecasting, 23(1),
Gallant, A.R. (2008). Nonlinear Statistical Models. New York. USA: John Wiley and Sons.
Gracia, C.J., Restrepo, J.E., and Roger, S. (2011). How Much Should Inflation Targeters Care About the Exchange Rate?, Journal of International Money and Finance 30(7),
150Buletin Ekonomi Moneter dan Perbankan, Volume 19, Nomor 2, Oktober 2016
Harding, D., and Pagan A. (2009). An Econometric Analysis of Some Models for Constructed Binary Time Series. NCER Working Paper 39.
Kaminsky, G., Lizondo, S., Reinhart, C. (1998). Leading Indicators of Currency Crises. IMF Staff papers, 45 (1),
Kauppi, H., Saikkonen, P. (2008). Predicting U.S. Recession with Dynamic Binary Response Models. The Review of Economics and Statistics, 90 (4),
Kumar, M., Moorthy, U., and Perraudin, W. (2003). Predicting Emerging Market Currency Crashes. Journal of Empirical Finance, 10,
Neely, C.J. (2001). The Practice of Central Bank Intervention: Looking Under the Hood. Federal Reserve Bank of St. Louis Review May/June,
Neely, C.J. (2001). Central Bank Authorities’ Believe about Foreign Exchange Intervention. Journal of International money and Finance, 27(1),
Early Warning System and Currency Volatility Management in Emerging Market 151
Appendix: Constrained Maximum Likelihood Estimation (Kauppi and Saikkkonen, 2008)
Recall the general form of the model in the case of a logistic distribution function . Following Kauppi and Saikkonen, the
initial value π0 is set to ,
being the sample mean of exogenous variables. The initial condition for the β vector of parameters is given by an OLS estimation, while the initial δ is set to 0. Moreover, since δ is an autoregressive parameter, a constrained maximum likelihood estimation must be implemented. Nevertheless, the same results can be reached in a faster and easier way, by using a transformation of the δ parameter in the classical maximum likelihood process. Thus, to solve this problem, denote the new maximization parameter by ψ, identified so that δ is equal to ψ/(1+|ψ|), i.e., δ takes value in the interval [0,1].
Hence, the
where θ is the vector of parameters θ = [ψ,α,β].
It is noticed that in view of the parameter transformation from δ to ψ, the maximization
estimated parameter , where
, the approximation becomes:
Nevertheless, we aim at finding the variance of δ, and thus, using the formula Var(a’ X) = a’ Var(X)a, we obtain:
152Buletin Ekonomi Moneter dan Perbankan, Volume 19, Nomor 2, Oktober 2016
Since , ψ0 can be replace with the estimator
Last but not least, the first derivative of the transformation function
with respect to can be computed through finite differences. Consequently, the standard error obtained as the square root of the elements laying on the first diagonal of the
can be obtained as the diagonal elements of the matrix , where
,
and where |
|
. |
|
On top of that, consider that the robust