Printer Friendly

Output recovery after currency crises.

INTRODUCTION

In December 2013, Ben Bernanke, chairman of the Board of Governors of the Federal Reserve System at the time, announced that the Federal Reserve would begin cutting back on its program of quantitative easing. The announcement induced downward pressure on the currencies of several emerging market economies, including Argentina, Turkey and South Africa. The central banks of these countries sold foreign exchange reserves or raised short-term interest rates to support the value of their currencies. Despite the support, these currencies were eventually devalued. (1)

The new wave of currency devaluations in emerging markets has renewed interest in some of the unanswered questions in the currency crises literature, including determining (a) whether currency crises entail output costs and (b) whether the output costs are permanent (L-shaped recovery) or temporary (V- or U-shaped recovery). The empirical literature concerning (a) primarily examines the short-run impact of currency crises on output and finds that such crises are often associated with output losses (see Cooper, 1971; Edwards, 1989; Gupta et al, 2007; Hutchison and Noy, 2002, 2005; Morley, 1992). The literature concerning (b) has received less attention in the theoretical and empirical literature.

The main objective of this paper is to explore the latter question. In particular, we examine the shape and long-run dynamics of output recovery following currency crises using a panel of 104 countries from 1960 to 2013. We apply the methodology proposed by Kim et al (2005), which is used to examine the post-recession recovery of output, to investigate the post-currency-crisis recovery of output. Specially, we estimate models that allow for U-, V-, and L-shaped responses of output to a currency crisis and allow the data to select the appropriate model. We then use the selected model to estimate the long-run path of output recovery following a currency crisis. A secondary objective of the paper is to determine whether output responds differently during currency crises that are characterized by a large devaluation of the nominal exchange rate versus those that are characterized by large reserve losses.

In general, the long-run dynamics and the shape of output recovery from currency crises are unsettled in both the theoretical and empirical literature. The theoretical literature develops three generations of models to explain the causes of currency crises. These models do not explicitly discuss the short- or long-run cost of crises. However, it is possible to infer the possible shape of output recovery (U-, V-, or L-shaped) from them. For example, in first and second generation models of currency crises, one key assumption that allows for contractionary outcomes is excess rigidities of nominal wages over prices (see Krugman and Taylor, 1978). The nominal nature of these rigidities implies that output should fully recover in the long run (that is, a V- or U-shaped recovery) (see Kamin and Rogers, 2000; Krugman and Taylor, 1978). Alternatively, third generation models of currency crises stress the role of financial fragility, moral hazard in the banking system, and over-lending to explain the occurrence of currency crises, which can imply a large permanent decline in output (L-shaped recovery) (see Aghion et al., 2001; Chang and Velasco, 2001; Diaz-Alejandro, 1985; McKinnon and Pill, 1995; Velasco, 1987).

The empirical literature on long-run cost of currency crises is sparse and the findings are mixed. For instance, Hong and Tornell (2005) examine all crisis episodes and compare post-crisis growth rates with the average growth rate in the tranquil years. Following currency crises, they find that the growth rate of output merely returns to its trend; thus, they conclude that output level grows on a permanently lower path (ie, L-shaped recovery). Using a similar methodology, Park and Lee (2001) report a V-shaped pattern of output recovery after the 1998 Asian financial crisis. The main shortcoming of these event study approaches is that they do not control for country fixed effects or global shocks.

Alternatively, Cerra and Saxena (2008) and Bussiere et al. (2012) use Autoregressive Distributed Lag models (ARDL) of growth and impulse responses to investigate the output costs associated with currency crises in the short and long run. This approach more appropriately allows them to control for country and time fixed effects. They find that currency crises lead to a permanent decline in the level of output (L-shaped recovery).

Our study extends the latter studies in two ways. First, the definition of currency crises in these studies combines both currency crashes and reserve crises together. We study these episodes separately.2 In addition, these studies consider only one possible shape of output recovery following a currency crisis. We consider several possible shapes and allow the data to select the appropriate model. Given the mixed empirical results in the literature, allowing several possible shapes of recovery is more appropriate.

Our empirical results suggest that the shape of output recovery from currency crashes is, on average, V-shaped. We find that currency crashes lead to large initial declines in output, but there is a significant bounce back in the growth rate of output following these events. Further examination of the long-run dynamics of output recovery shows that the bounce back allows output to return to its non-crisis level within two to three years. We also find that reserve crises are associated with a smaller initial output loss, but there is a drag in growth rate of output following a crisis. This results in a larger and more persistent output loss in the long run. A possible channel through which reserve crises can impact output is the reduction in the monetary base leading to a credit crunch and loss of output. (3)

Our findings suggest that defending a currency by using foreign reserves is costly in the long run. In addition, the results indicate that the mixed findings of previous work concerning the shape and dynamics of output recovery from currency crises is at least partially due to the use of a measure of currency crisis that includes both reserve crises and currency crashes.

DATA DESCRIPTION AND EMPIRICAL MODEL

Data

We obtained the available annual macroeconomic data for 104 countries for the years 1960-2013 from the World Bank's World Development Indicators (WDI) and International Financial Statistics CD-ROM (World Bank, 2014). The variables we employed include real GDP in constant 2000 US dollars, which is used to calculate the annual growth rate of output, the official nominal exchange rate (LCU per US$, period average) and non-gold international reserves, which are used to identify currency crashes and reserve crises episodes, respectively.

In order to reconcile some of the differences between previous studies, we employ several measures of currency crises.4 The common practice in the empirical literature is to construct an Exchange Market Pressure Index (EMPI) to capture episodes of severe pressure on the foreign exchange market. This index is the weighted average of the depreciation of the nominal exchange rate, the percentage change in the foreign reserves, and the change in the interest rate (eg see Eichengreen et al, 1995). Currency crises are then defined as episodes in which EMPI is greater than some threshold.5 Competing studies select different components of the EMPI (and thresholds) to identify crises. Most studies exclude the interest rate change component due to the lack of reliable market-oriented interest rate data. Other studies also exclude the percentage change in foreign reserves and focus on episodes of currency crashes. The currency depreciation component of EMPI captures all episodes of severe pressure on the foreign exchange market that led to a large depreciation (successful attacks). However, the other two components of EMPI capture episodes in which the central bank intervened in the market to defend its currency (defense episodes).

We suspect that the type of currency crisis might influence the shape of output recovery, so we follow Aziz et al. (2000) and distinguish between 'reserve crises' and 'currency crashes'. (6) In order to identify currency crashes we employ the methodology of Laeven and Valencia (2008), who followed Frankel and Rose (1996), and define a crisis as: 'a nominal depreciation of at least 30% that is at least 10% increase in the rate of depreciation compared to the year before'. (7) Additionally, if countries meet the criteria for several continuous years, Laeven and Valencia (2008) only identify the first year of each 5-year window as a crisis. Their definition yields 183 currency crises during the period 1960-2013 in our sample (labeled LV). However, the availability of growth rate data reduces the number of crises with complete data to 170 episodes.

We are interested in examining how the recovery depends on the duration of crises; therefore, we need to account for the serial correlation of the crisis dummy variable. Hence, we construct our dummy variable labeled EVENT the same way as LV above, but with no window imposed around each crisis. Therefore, if countries meet the criteria for several continuous years, the EVENT dummy identifies all the years as crisis years. Our EVENT dummy includes 287 country-years of crises for which GDP growth rates are available.

To identify reserve crises, we first follow Cerra and Saxena (2008) characterization of crises. They construct an EMPI, which is based on the percentage depreciation in the nominal exchange rate plus the percentage loss in foreign exchange reserves. Then they form a dummy variable, labeled CS (Table 1(a)), that is equal to one any time EMPI is in the upper quartile of all observations across the entire panel. The dummy variable is equal to zero otherwise. Taking into account the availability of GDP growth, the criteria yield 1,231 episodes of crises in our sample. We take this variable and, similar to Aziz et al. (2000), break it into two other dummy variables: CRISIS and CRISISR.

The CRISIS dummy variable provides an alternative way to characterize currency crashes and enables us to further check the robustness of our results. This variable is equal to one if the exchange rate component of EMPI accounts for at least 75% of the index when the index signals a crisis (when CS = 1). Otherwise it is zero. (8) There are some differences between the episodes captured by the CRISIS and EVENT dummy. For example the CRISIS dummy excludes all episodes in which depreciation is large but the loss in foreign exchange reserve is much larger, where the EVENT dummy includes these episodes as crises. (9)

Finally, we construct the CRISISR dummy variable similar to CRISIS variable only that the foreign exchange reserves component of EMPI must account for at least 75% of the index when the index signals a crisis (when CS = 1). (10) We end up with 297 and 567 episodes of currency crashes and reserve crises, respectively. Table 1 (b) compares CRISIS and CRISISR dummies with our EVENT dummy variable.

All countries included in this analysis have experienced a currency crisis at some point during the sample period. We exclude those countries that have never experienced a speculative attack during our sample period because these countries may be inherently different from other countries and this can lead to bias in our results (see Aziz et al, 2000; Frankel and Rose, 1996; Hong and Tornell, 2005). Therefore, the non-crisis (tranquil) episodes comprise the control group. (11)

Related literature and empirical model

Our empirical strategy is borrowed from the literature on output recovery after recessions; therefore, we begin by briefly discussing this literature. The models used to examine post-recession recoveries are based on the observation that output's behavior is asymmetric following expansions and contractions (see Beaudry and Koop, 1993; Hamilton, 1989; Luukkonen and Terasvirta, 1991; Neftci, 1984). Hamilton (1989) was the first to model this nonlinearity using a Markov switching model of GDP growth rate. Hamilton's model is an autoregressive model of GDP growth rate in which the mean of GDP growth depends on the state of economy (recession or expansion), which is endogenously estimated. The empirical evidence based on Hamilton's model documents a relatively large and permanent output loss for post-war United States (L-shaped recovery). However, follow-up research revealed that the estimates based on Hamilton's model may be too large because his model fails to account for a recovery phase in which output's growth may grow at a rate higher than the typical growth rate. In order to account for this phase, Sichel (1994) added another regime to Hamilton's model. Beaudry and Koop (1993) augment a standard ARMA model of output growth with a current-depth-of recession dummy variable and found it to be significant and positive.

Alternatively, Kim et al. (2005) (see also Bee et al, 2014; Morley and Piger, 2012) augments Hamilton's autoregressive model of output's growth rate to include a bounce back term that is proportional to the length or depth of recession. Hence, they propose three main types of bounce back functions, which correspond to a U-shaped, V-shaped, and finally a Depth bounce back model which allows for nonlinear recoveries of output depending on the depth of a recession. Thereafter, based on the information criteria they select between these alternatives and Hamilton's original model, which represents an L-shaped recovery of output.

Following Kim et al. (2005) and Morley and Piger (2012) and using a panel of 104 countries over the 1960-2013 period, we estimate the U-, V- and L-shaped models of recovery in the context of currency crises. Unlike the studies that estimate the timing of a recession endogenously, we identify our regime switching with a more deterministic approach based on different identification schemes for currency crises we previously described. Thereafter, based on the information criteria, we allow the data to select the appropriate model.

This approach has several advantages. First, unlike the event study methodology used in the literature, our approach allows us to control for country and time fixed effects. Second, we are able to test for the significance of a bounce back term and the possibility of a full-recovery in the level of output in the long run. Third, unlike Cerra and Saxena (2008), which also examines the long-run output costs associated currency crises, we cover a broader range of possibilities for output recovery. We allow the information criteria to select between our model and that of Cerra and Saxena (2008) using different measures of a currency crisis. Finally, we are able to calculate and reevaluate the long-run response of output to a currency crisis based on our selected model. Thus, we estimate the following bounce back models of growth within a panel data framework:

[phi](L)([DELTA][y.sub.it] - [[mu].sub.it]) = [[epsilon].sub.it]

where:

(1) Hamilton's Model: L-shaped recovery:

[[mu].sub.it] = [[alpha].sub.i] + [[theta].sub.t] + [[gamma].sub.1] [D.sub.it]

(2) U-shaped recovery:

[[mu].sub.it] = [[alpha].sub.i] + [[theta].sub.t] + [lambda] [D.sub.it] + [lambda] [m.summation over (j=1)] [[gamma].sub.1] [D.sub.i,t-j]

(3) V-shaped recovery:

[[mu].sub.it] = [[alpha].sub.i] + [[theta].sub.t] + [[gamma].sub.1] [D.sub.it] + (1 - [D.sub.it]). [lambda] [m.summation over (j=1) [[gamma].sub.1] [D.sub.i,t-j]

(4) Cerra and Saxena Model:

[[mu].sub.it] = [[alpha].sub.i] + [[theta].sub.t] + [4.summation over (j=0)] [[gamma].sub.j+1] [D.sub.it-j]

where [DELTA][y.sub.i] is the annual growth rate of real GDP in country i, [phi](L) is a lag polynomial of order p with roots lying outside the unit circle, [D.sub.it] is the currency crisis dummy for country i in year t, [[alpha].sub.i], is a country-specific fixed effect, and 6t is time-specific fixed effect. The m parameter represents the duration of the bounce back term.

While there are an infinite number of possibilities for modeling growth dynamics, these models cover a broad range of possibilities, allowing us to capture whether the impact of a currency crisis on the level of output is permanent or temporary. Based on previous literature we would expect [[gamma].sub.1] (the coefficient of contemporaneous dummy for currency crisis) to be negative, meaning that a currency crisis is associated with a decline in the growth rate of output. Therefore, a bounce back takes place if [lambda] is negative and statistically significant. In this case, the last term in all the above equations will be positive. This implies that the growth rate of output will be greater than non-crisis average ([[theta].sub.t] + [[alpha].sub.i]) in the m years following a crisis which allows output to level off if [lambda] is large enough.

Note that though these models focus on the dynamics of growth rates, they will tell us about the shape of output's level recovery. Figure 1 illustrates this by presenting the path of output following a crisis under the three models above when m = 4 and [lambda] = -0.3. The V-shape recovery of model 3 is different from the other bounce back function U in that, due to the (1 - [D.sub.it]) term, the bounce back term kicks in only when the crisis is completely over and [D.sub.it] is zero, thus giving output recovery the V-shape.

In order to be consistent with Cerra and Saxena (2008), we choose L- 4 (autoregressive lags) and m = 4 (duration of bounce back term). This (m = 4) means that we allow for the bounce back term to generate a higher than non-crisis growth rate for four years after the crisis.

For ease of presentation and understanding, we present the coefficients on all dummy variables and estimated [lambda] in the tables. However, when interpreting the results we use the composite of [[gamma].sub.1] and [lambda], equal to [lambda]*[[gamma].sub.1]. This magnitude represents the annual percentage point bounce back from the initial decline. In the case of model 3 (V-shaped model) this coefficient doesn't turn on until the crisis is complete.

SHAPE OF RECOVERY AFTER CURRENCY CRISES

In this section, we estimate equation 1 through 4 using the EVENT dummy variable and report the Akaike (AIC) and Schwarz (SIC) information criterion corresponding to each model in Table 2. As previously mentioned, in order to control for fixed country-specific effects and common global shocks across time, each model is estimated using a panel two-way fixed effects method. (12) In Table 2 we find that both the AIC and SIC select the model with a V-shaped recovery. Estimation results indicate a decline of between 1.6 and 2.3 percentage points in the growth rate of output during a crisis. The bounce back term implies that a year after a crisis, the average growth rate gets a boost of about 1.01 (equal to [lambda]*[[gamma].sub.1] = - 0.628* - 1.623) percentage point per year after the crisis which is scaled by the duration of each crisis. This bounce back can be the result of a boom in net export or/and reduction in imports following a large devaluation. Therefore, the results suggest that the growth rate of output in the recovery phase is in fact higher than the non-crisis period growth rate.

In addition, the results thus far indicate that the V-shaped model is preferred over the model proposed by Cerra and Saxena (2008). Next, we use the parameter estimates of our preferred model (V-shaped) and following Cerra and Saxena (2008) calculate the impulse response of output to a currency crisis characterized by EVENT. For example, the initial response of output ([partial derivative][y.sub.it]/[partial derivative][D.sub.it]) to a currency shock will be [[gamma].sub.1]. The 1-year ahead cumulative response will be [[gamma].sub.1] + ([[phi].sub.1].[[gamma].sub.1]+[lambda].[[gamma].sub.1]), where [[phi].sub.1] is the autoregressive coefficient. We assess the significance of the results by computing one standard deviation confidence bands derived from 1,000 Monte-Carlo simulations. We refer to responses as statistically significant if the confidence bands exclude the zero line.

Calculating impulse responses allows us to evaluate long-run output costs based on our preferred model and to compare it with that of Cerra and Saxena (2008). The negative and significant [lambda] tells us that there is a bounce back; however, we are also interested in seeing if the bounce back is large enough to guarantee a complete recovery. The impulse response of output's level enables us to see whether output fully recovers to its non-crisis trend.

We present the impulse response in the left panel of Figure 2 and it suggests that output fully recovers to its non-crisis level within two years. This is in line with the findings of Park and Lee (2001), but in direct contrast to Cerra and Saxena's (2008) results who find a permanent decline of about 4% points in the level of output following a currency crisis.

Currency crises are often accompanied by other types of financial crises such as debt and banking crises. Therefore, our estimation of the output costs can be overstated by attributing all output losses to currency crises. In order to address this issue, we obtain banking and debt crisis dates from Laeven and Valencia (2008) and construct a dummy variable corresponding to each of these events. We then augment model 1 through 4 with a banking crisis and a debt crisis dummy variables and re-estimate each model. The coefficient of the contemporaneous currency crisis dummy variable, [[gamma].sub.1], now represents the initial impact of a pure currency crisis on output growth.

The results, presented in Table 3, show the selected model is still V-shaped. However, the initial decline in output growth is now smaller (about 1 percentage point) in all models. Moreover, the bounce back term is larger (-1.65*-0.922 = 1.074). This implies that absence of any other type of financial turbulence, currency crisis, characterized by large nominal exchange rate devaluation, are associated with a smaller output losses and a larger output rebound. Using the V-shape model (the preferred model) we calculate the impulse response of output to a pure currency crisis shock and present it in the right panel of Figure 2.

Besides the different specification of the models, the difference between our results and those of Cerra and Saxena (2008) can be due to different characterizations of episodes of currency crises. Their definition of crisis includes both currency crashes and reserve crises episodes. Our characterization of currency crises thus far, only includes currency crashes. To further examine the association between the shape of recovery and type of currency crises, we use the two dummy variables we constructed earlier (CRISIS and CRISISR) and estimate equations 1-4 once again.

The result suggests that the selected model of recovery from currency crashes, characterized by CRISIS, is again the V-shaped model (Table 4). The bounce back term is significant and in terms of magnitude it is similar to the bounce back term in the EVENT episodes. Augmenting the regressions to include debt and banking crisis dummy variables does not qualitatively change the results. (13) The results for CRISISR are, however, interesting and mixed. Table 5 shows the estimation results with the CRISISR dummy.

Table 5 indicates that according to SIC the recovery after reserve crises is U-shaped, but Cerra and Saxena's (2008) model is preferred by the AIC. It is worth noting that models selected by SIC tend to be more parsimonious relative to those selected by AIC. The interesting result is that lambda is positive and significant, indicating that there is no bounce back in growth rates following reserve crises. In fact, there is a drag term that dampens output growth. Additionally, the initial output loss is much smaller in case of a reserve crisis.

To further compare the results with that of Cerra and Saxena (2008), we estimate the model proposed by Cerra and Saxena (model 4) with CRISIS and CRISISR. Thereafter, we calculate the impulse response of the level of output to these events, separately. The impulse responses are presented in Figure 3.

It is clear that while currency crashes are initially more costly, in the medium run (within three years) output fully recovers to its non-crisis level. On the other hand, reserve crises are associated with a small decline in output in the short run, but a large (5% to 6%) loss in the level of output in the long run. These results indicate that Cerra and Saxena (2008) finding of permanent loss in the level of output following currency crises is mainly driven by reserve crises rather than episodes of currency crashes.

The findings are similar when we augment model 4 with the banking and debt crisis dummy variables and re-calculate the impulses responses of output to a pure currency crash and pure reserve crises (see Figure 4).

The impulse responses show that output fully recovers following currency crashes that are not accompanied by any type of other financial crises. (14) In terms of policy implications, our results indicate that policy makers face a trade-off. Although defending the currency in the short run by depleting foreign exchange reserves entails a small output cost and seems like a better option than allowing the currency to collapse, it is more costly in the long run. Hutchison et al. (2010) find similar results in the case of sudden stops. Their results suggest that contractionary fiscal and monetary policy (an increase in the discount rate or an unsterilized decline in international reserves) are associated with significantly larger output losses after sudden stops. Moreover, Erler et al. (2014) find that in face of speculative attacks, immediate depreciation or a successful defense of a currency entail smaller output costs than an unsuccessful intervention.

How can we explain the decline in output after a currency defense? In order to defend the currency under a speculative attack, the central bank has to engage in unsterilized intervention in the foreign exchange market (selling foreign reserves). Unsterilized interventions lead to declines in the monetary base and increases in the short-term domestic interest rates. (15) Higher interest rates entail three types of costs. First, they increase the fiscal cost associated with public debt. Second, they aggravate the asymmetric information problem in the financial markets by making it more difficult for lenders to distinguish good from bad borrowers; thus, it leads to a credit crunch and eventual output loss. Finally, they can deteriorate the balance sheet of a weak banking system leading to bankruptcies and a reduction in output (Lahiri and Vegh, 2007).

ROBUSTNESS CHECK

In some cases depleting reserves to defend a currency against devaluation only postpones devaluation (unsuccessful defense). Erler et al. (2014) evaluate the short-run costs associated with currency crashes, successful defenses, and unsuccessful defenses and find that unsuccessful defenses are the most costly. We do not distinguish between successful and unsuccessful defense in our analysis. Therefore, our results on the output cost of reserve crises (defense) can be due to the fact that some reserve crises were followed by large devaluations; hence they entailed larger output loss. In order to check for this, we exclude all episodes of reserve crises that proceeded large devaluations by one or two years and re-estimate the impulse responses. (16) The results do not qualitatively change; reserve crises that are not followed by a large devaluation (successful defenses) are still costly in the long run (see Figure 5).

In addition, as pointed out by Teulings and Zabanov (2014), the impulse response functions (IRFs) derived from an ARDL model of output growth and crisis dummy variables (called analytical IRF) is sensitive to the choice of number of lags and thus unstable. (17) They follow Jorda's (2005) suggestion of using local projections to estimate IRFs. The IRF estimated using this method is robust to misspecification of the data-generating process, because instead of using the same set of coefficient for all forecasting horizons, it uses a new set of coefficients for each time horizon (Jorda, 2005; Teulings and Zabanov, 2014). The disadvantage of this method is that the IRF estimates are less efficient.

In this paper, we chose to follow Cerra and Saxena (2008) and calculate analytical IRFs instead of local projections IRFs, because it makes it possible to compare our results with that of Cerra and Saxena (2008). However, we check the robustness of our results first: (i) by choosing different lags of crisis dummy and autoregressive lags in the ARDL model (Figure 6); (ii) by estimating local projections IRFs for both CRISIS and CRISISR (Figure 7).

In Figure 6, we present the analytical IRFs with different lags of the CRISIS dummy and autoregressive lags. The impulse responses indicate that currency crashes are more costly in the short run and reserve crises are more costly in the long run. It is worth noting that the choice of numbers of lags of the CRISIS dummy variable proves to be important for the long-run recovery. Inclusion of higher lags of the CRISIS dummy in the estimation indicates a full recovery of output in the long run following a currency crash while excluding the higher lags implies a permanent decline in the level of GDP. However, the decline in GDP is still lower in this case than in the case of reserve crises. Therefore, our main conclusion that reserve crises are more costly in the long run still remains valid. Figure 7 shows that the results are similar when we estimate local projections IRFs using CRISIS and CRISISR episodes. (18)

Finally, the models are estimated assuming that currency crises are exogenous to output growth. However, it is possible that currency crises are the result rather than the cause of the decline in output and negative growth. This possibility has been rigorously examined in Cerra and Saxena (2008). They find suggestive evidence of growth optimism prior to financial crises, meaning that forecasts of economic growth before crises tend to be higher than actual growth. In order to address the endogeneity of currency crises we follow a simpler methodology put forth by Furceri and Zdzienicka (2012a, b) and include a recession dummy in our specifications of model 1-4. The recession dummy is equal to one when contemporaneous growth is negative after a period of positive growth (growtht < 0, [growths.sub.t-1] > 0) and zero otherwise. The results are qualitatively similar to our previous results; the recovery is still V-shaped and currency reserves are still more costly than currency crashes in the long run (Figure 8).

CONCLUSION

The theoretical literature on currency crises does not predict a particular pattern of output recovery following these events and the empirical literature is mixed. In this paper, we examine the shape and long-run dynamics of output recovery from currency crises in a panel of 104 countries during 1960-2013. Specially, we estimate U-, V-, and L-shaped paths of output recovery after a currency crisis and allow the data to select the appropriate model.

In addition, we suspect that the mixed empirical findings in the literature regarding the shape and long-run output cost following currency crises is due to different types of currency crises considered. Therefore, in our analysis, we distinguish between two major types of currency crises, namely currency crashes and reserve crises (currency defense episodes).

Our results indicate that output recovery following currency crashes is, on average, V-shaped. The long-run analysis revealed that following currency crashes, output initially collapses; however, it fully recovers to its non-crisis average within two to three years. On the other hand, reserve crises entail small initial output losses, but in the long run

the output loss is large (5% to 6%) and persistent. These results indicate that policy makers should take into account the long-run costs when deciding whether to allow devaluation of their currency or defending it. In addition, our results suggest the need for further research on the long-run costs associated with defending a currency.

APPENDIX

See Tables A1-A4.

Table A1: Model estimates using CRISIS and
controlling for other types of financial crises

Dependent variable:         (1)              (2)
GDP growth                L-shaped         U-shaped

[DELTA][y.sub.i,t-1]        0.216 ***        0.216 ***
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-2]        0.008            0.010
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-3]        0.017            0.019
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-4]        0.000            0.004
                           (0.016)          (0.016)
[CRISIS.sub.t]             -1.759 ***       -1.976 ***
  ([[gamma].sub.1])        (0.424)          (0.431)
Banking crisis             -2.026 ***       -2.115 ***
                           (0.426)          (0.427)
Debt crisis                -3.094 ***       -2.902 ***
                           (0.913)          (0.915)
[lambda]                                    -0.249 ***
                                            (0.090)
[CRISIS.sub.t_1]

[CRISIS.sub.t_2]

[CRISIS.sub.t_3]

[CRISIS.sub.t_4]

Constant                    3.565 ***        2.713 ***
                           (0.991)          (0.991)
SIC                    22,087.0         22,087.2
AIC                    21,755.5         21,749.5
N                       3,425            3,425
[R.sup.2]                   0.127            0.129

Dependent variable:          (3)                (4)
GDP growth                V-shaped       Cerra and Saxena

[DELTA][y.sub.i,t-1]        0.216 ***         0.215 ***
                           (0.017)           (0.017)
[DELTA][y.sub.i,t-2]        0.009             0.010
                           (0.017)           (0.017)
[DELTA][y.sub.i,t-3]        0.019             0.019
                           (0.017)           (0.017)
[DELTA][y.sub.i,t-4]        0.005             0.004
                           (0.016)           (0.016)
[CRISIS.sub.t]             -1.341 ***        -1.944 ***
  ([[gamma].sub.1])        (0.444)           (0.433)
Banking crisis             -2.055 ***        -2.110 ***
                           (0.426)           (0.428)
Debt crisis                -2.982 ***        -2.920 ***
                           (0.913)           (0.916)
[lambda]                   -0.477 ***
                           (0.152)
[CRISIS.sub.t_1]                              0.206
                                             (0.433)
[CRISIS.sub.t_2]                              0.533
                                             (0.425)
[CRISIS.sub.t_3]                              0.632
                                             (0.422)
[CRISIS.sub.t_4]                              0.569
                                             (0.413)
Constant                    3.446 ***         3.465 ***
                           (0.990)           (0.991)
SIC                    22,084.8#         22,111.0
AIC                    21,747.2#         21,754.9
N                       3,425             3,425
[R.sup.2]                   0.129             0.129

Note: The dependent variable is the growth rate of GDP
([DELTA][y.sub.it]). The [[gamma].sub.1] represents the coefficient
of the simultaneous crisis dummy (CRISIS) and [lambda] *
[[gamma].sub.1] represents the magnitude of drag in output growth.
Banking crisis and Debt crisis are dummies, which are equal to one
if there is a banking or debt crisis in a country in a year. The
dates are taken from Laeven and Valencia (2008) Sample includes 104
countries during the 1960-2013 period. Country and year fixed
effects are both included and the estimation method is two-way
fixed. AIC and SIC correspond to Akaike and Schwarz criterion,
respectively. The lower values of AIC and SIC represent the
preferred model (highlighted in bold). *, **, and *** represent
significant at 1%, 5%, and 10%, respectively. The reported
[R.sup.2] is within [R.sup.2].

Note: The lower values of AIC and SIC represent the
preferred model (highlighted in bold) indicated with #.

Table A2: Model estimates using CRISIS[R.sup.2]

Dependent variable:         (1)              (2)
GDP growth                L-shaped         U-shaped

[DELTA][y.sub.i,t-1]        0.156 ***        0.151 ***
                           (0.016)          (0.016)
[DELTA][y.sub.i,t-2]        0.062 ***        0.060 ***
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-3]        0.038 **         0.037 **
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-4]        0.010            0.009
                           (0.016)          (0.016)
[CRISISR.sub.t]            -0.704 *         -0.705 *
  ([[gamma].sub.1])        (0.362)          (0.361)
[lambda]                                     1.108 ***
                                            (0.251)
[CRISISR.sub.t_1]

[CRISISR.sub.t_2]

[CRISISR.sub.t_3]

[CRISISR.sub.t_4]

Constant                    2.655 ***        3.209 ***
                           (1.028)          (1.033)
SIC                    24,046.9         24,034.8#
AIC                    23,711.6         23,693.3#
N                       3,674            3,674
[R.sup.2]                   0.087            0.092

Dependent variable:         (3)               (4)
GDP growth                V-shaped      Cerra and Saxena

[DELTA][y.sub.i,t-1]        0.152 ***        0.151 ***
                           (0.016)          (0.016)
[DELTA][y.sub.i,t-2]        0.061 ***        0.061 ***
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-3]        0.037 **         0.036 **
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-4]        0.010            0.008
                           (0.016)          (0.016)
[CRISISR.sub.t]            -1.173 ***       -0.666 *
  ([[gamma].sub.1])        (0.384)          (0.362)
[lambda]                    0.587 ***
                           (0.162)
[CRISISR.sub.t_1]                           -1.241 ***
                                            (0.361)
[CRISISR.sub.t_2]                           -0.322
                                            (0.363)
[CRISISR.sub.t_3]                           -0.575
                                            (0.362)
[CRISISR.sub.t_4]                           -1.024 ***
                                            (0.358)
Constant                    3.176 ***        3.161 ***
                           (1.036)          (1.033)
SIC                    24,041.4         24,055.4
AIC                    23,700.0         23,695.3
N                       3,674            3,674
[R.sup.2]                   0.090            0.093

Note: The dependent variable is the growth rate of GDP
([DELTA][y.sub.it]). The [[gamma].sub.1] represents the coefficient
of the contemporaneous reserve crisis dummy (CRISISR2). The dummy
represents all episodes of reserve crises that were not followed by
a large depredation (15% or more) of nominal exchange rate in the
next two years. [lambda] * [[gamma].sub.1] represents the magnitude
of drag in output growth. Sample is based on available data during
the 1960-2013 period for 104 countries. Country and year fixed
effects are both included and the estimation method is two-way
fixed. AIC and SIC correspond to Akaike and Schwarz criterion,
respectively. The lower values of AIC and SIC represent the
preferred model (highlighted in bold). *, **, and *** represent
significant at 1%, 5%, and 10%, respectively. The reported
[R.sup.2] is within [R.sup.2].

The lower values of AIC and SIC represent the preferred model
(highlighted in bold) indicated with #.

Table A3: Model estimates using event dummy and controlling for
recessionary episodes

Dependent variable:         (1)              (2)
GDP growth                L-shaped         U-shaped

[DELTA][y.sub.i,t-1]        0.258 ***        0.257 ***
                           (0.014)          (0.014)
[DELTA][y.sub.i,t-2]        0.064 ***        0.067 ***
                           (0.015)          (0.015)
[DELTA][y.sub.i,t-3]        0.044 ***        0.047 ***
                           (0.015)          (0.015)
[DELTA][y.sub.i,t-4]       -0.001            0.003
                           (0.014)          (0.014)
[EVENT.sub.t]              -1.350 ***       -1.419 ***
  ([[gamma].sub.1])        (0.389)          (0.389)
[D.sup.g]                  -8.707 ***       -8.648 ***
                           (0.316)          (0.315)
[lambda]                                    -0.469 ***
                                            (0.128)
[EVENT.sub.t_1]

[EVENT.sub.t_2]

[EVENT.sub.t_3]

[EVENT.sub.t_4]

Constant                    3.895 ***        3.756 ***
                           (0.770)          (0.770)
SIC                    26,250.8         26,245.2
AIC                    25,903.5         25,891.7
N                       4,079            4,079
[R.sup.2]                   0.275            0.277

Dependent variable:         (3)               (4)
GDP growth                V-shaped      Cerra and Saxena

[DELTA][y.sub.i,t-1]        0.256 ***        0.253 ***
                           (0.014)          (0.014)
[DELTA][y.sub.i,t-2]        0.066 ***        0.065 ***
                           (0.015)          (0.015)
[DELTA][y.sub.i,t-3]        0.048 ***        0.048 ***
                           (0.015)          (0.015)
[DELTA][y.sub.i,t-4]        0.003            0.005
                           (0.014)          (0.014)
[EVENT.sub.t]              -0.905 **        -1.414 ***
  ([[gamma].sub.1])        (0.404)          (0.389)
[D.sup.g]                  -8.650 ***       -8.648 ***
                           (0.315)          (0.315)
[lambda]                   -0.858 ***
                           (0.214)
[EVENT.sub.t_1]                             -0.231
                                            (0.385)
[EVENT.sub.t_2]                              0.628 *
                                            (0.381)
[EVENT.sub.t_3]                              1.056 ***
                                            (0.379)
[EVENT.sub.t_4]                              1.186 ***
                                            (0.377)
Constant                    3.725 ***        3.817 ***
                           (0.770)          (0.771)
SIC                    26,242.5#        26,261.7
AIC                    25,888.9#        25,889.2
N                       4,079            4,079
[R.sup.2]                   0.277            0.278

Note: The dependent variable is the growth rate of GDP
([DELTA][y.sub.it]). The [[gamma].sub.1] represents the coefficient
of the simultaneous crisis dummy (EVENT). [lambda] *
[[gamma].sub.1] represents the magnitude of bounce back or drag in
output growth. Banking crisis and Debt crisis are dummies, which
are equal to one if there is a banking or debt crisis in a country
in a year. The Recession dummy is a dummy that is equal to one if
([growth.sub.t] < 0, [growth.sub.t-1] > 0) and zero otherwise. The
sample includes 104 countries during the 1960-2013 period. Country
and year fixed effects are both included and the estimation method
is two-way panel fixed effects. AIC and SIC correspond to Akaike
and Schwarz criterion, respectively. The lower values of AIC and
SIC represent the preferred model (highlighted in bold). *, **, and
*** represent significant at 1%, 5%, and 10%, respectively. The
reported [R.sup.2] is within [R.sup.2].

The lower values of AIC and SIC represent the preferred
model (highlighted in bold) indicated with #.

Table A4: Model estimates using CRISISR dummy
and controlling for recessionary episodes

Dependent variable:         (1)              (2)
GDP growth                L-shaped         U-shaped

[DELTA][y.sub.i,t-1]        0.188 ***        0.183 ***
                           (0.015)          (0.015)
[DELTA][y.sub.i,t-2]        0.047 ***        0.044 ***
                           (0.016)          (0.016)
[DELTA][y.sub.i,t-3]        0.021            0.019
                           (0.016)          (0.016)
[DELTA][y.sub.i,t-4]        0.005            0.005
                           (0.015)          (0.015)
[CRISISR.sub.t]            -0.693 **        -0.714 **
  ([[gamma].sub.1])        (0.289)          (0.288)
Recession dummy            -8.528 ***       -8.463 ***
                           (0.324)          (0.324)
[lambda]                                     0.761 ***
                                            (0.202)
L. CRISISR

L2. CRISISR

L3. CRISISR

L4. CRISISR

Constant                    4.801 ***        5.247 ***
                           (0.952)          (0.958)
SIC                    22,907.6         22,901.0#
AIC                    22,567.2         22,554.4
N                       3,602            3,602
[R.sup.2]                   0.236            0.239

Dependent variable:         (3)               (4)
GDP growth                V-shaped      Cerra and Saxena

[DELTA][y.sub.i,t-1]        0.185 ***        0.183 ***
                           (0.015)          (0.015)
[DELTA][y.sub.i,t-2]        0.045 ***        0.047 ***
                           (0.016)          (0.016)
[DELTA][y.sub.i,t-3]        0.019            0.018
                           (0.016)          (0.016)
[DELTA][y.sub.i,t-4]        0.006            0.003
                           (0.015)          (0.015)
[CRISISR.sub.t]            -1.037 ***       -0.694 **
  ([[gamma].sub.1])        (0.316)          (0.288)
Recession dummy            -8.477 ***       -8.451 ***
                           (0.324)          (0.324)
[lambda]                    0.404 ***
                           (0.152)
L. CRISISR                                  -1.245 ***
                                            (0.287)
L2. CRISISR                                 -0.146
                                            (0.287)
L3. CRISISR                                 -0.130
                                            (0.287)
L4. CRISISR                                 -0.667 **
                                            (0.284)
Constant                    5.173 ***        5.198 ***
                           (0.962)          (0.957)
SIC                    22,908.4         22,914.8
AIC                    22,561.8         22,549.6#
N                       3,602            3,602
[R.sup.2]                   0.238            0.242

Note: The dependent variable is the growth rate of GDP
([DELTA][y.sub.it]). The [[gamma].sub.1] represents the coefficient
of the contemporaneous crisis dummy (CRISISR). [lambda] *
[[gamma].sub.1] represents the magnitude of bounce back or drag in
output growth. Banking crisis and Debt crisis are dummies, which
are equal to one if there is a banking or debt crisis in a country
in a year. The Recession dummy is a dummy that is equal to one if
([growth.sub.t] < 0, [growth.sub.t-1] > 0) and zero otherwise. The
sample includes 104 countries during the 1960-2013 period. Country
and year fixed effects are both included and the estimation method
is two-way fixed. AIC and SIC correspond to Akaike and Schwarz
criterion, respectively. The lower values of AIC and SIC represent
the preferred model (highlighted in bold). *, **, and *** represent
significant at 1%, 5%, and 10%, respectively. The reported
[R.sup.2] is within [R.sup.2].

The lower values of AIC and SIC represent
the preferred model (highlighted in bold) indicated with #.


REFERENCES

Aghion, P, Bacchetta, P and Banerjee, A. 2001: Currency crises and monetary policy in an economy with credit constraints. European Economic Review 45(7): 1121-1150.

Angkinand, A, Li, J and Willett, T. 2006: Measures of currency crises: A survey. Claremont Working Papers 2006-06. Claremont Graduate University: Claremont, CA.

Aziz, J, Caramazza, F and Salgado, R. 2000: Currency crises: In search of common elements. IMF Working Paper WP/00/67. International Monetary Fund: Washington DC.

Bee, F, Bouabdailah, O and Ferrara, L. 2014: Comparing the shape of recoveries: France, the UK and the US. Economic Modeling 44(c): 327-334.

Basistha, A and Teimouri, S. 2015: Currency crises and output dynamics. Open Economies Review 26(1): 139-153.

Beaudry, P and Koop, G. 1993: Do recessions permanently change output? Journal of Monetary Economics 31(2): 149-163.

Bussiere, M, Saxena, SC and Tovar, CE. 2012: Chronicle of currency collapses: Re examining the effects on output. Journal of International Money and Finance 31(4): 680-708.

Cerra, V and Saxena, S. 2008: Growth dynamics: The myth of economic recovery. American Economic Review 98(1): 493-457.

Chang, R and Velasco, A. 2001: A model of financial crises in emerging markets. Quarterly Journal of Economics 116(2): 489-517.

Cooper, RN. 1971: Currency devaluation in developing countries. In: Ranis, G and Ranis, G (eds). Government and economic development. Yale University Press: New Haven, CT.

Diaz-Alejandro, C. 1985: Good-bye financial repression, hello financial crash. Journal of Development Economics 19(1-2): 1-24.

Eichengreen, B, Rose, AK and Wyplosz, C. 1995: Exchange market mayhem: The antecedents and aftermath of speculative attacks. Economic Policy 10(21): 249-312.

Edwards, S. 1989: Real exchange rates, devaluation, and adjustment. MIT Press: Cambridge, MA.

Erler, A, Bauer, C and Herz, B. 2014: Defending against speculative attacks - It is risky, but it can pay off. Journal of Banking and Finance 47(c): 309-330.

Frankel, JA and Rose, AK. 1996: Currency crashes in emerging markets: An empirical treatment. Journal of International Economics 41(3-4): 351-366.

Furceri, D and Zdzienicka, A. 2012a: Banking crises and short and medium term output losses in emerging and developing countries: The role of structural and policy variables. World Development 40(12): 2369-2378.

Furceri, D and Zdzienicka, A. 2012b: The consequences of banking crises for public debt. International Finance 15(3): 289-307.

Gupta, P, Deepak, M and Ratna, S. 2007: Behavior of output during currency crises. Journal of International Economics 72(2): 428-450.

Hamilton, J. 1989: A new approach to the economic analysis of nonstationary time series and the business cycle. Econometrica 57(2): 357-384.

Hong, K and Tornell, A. 2005: Recovery from a currency crisis: Some stylized facts. Journal of Development Economics 76(1): 71-96.

Hutchison, M and Noy, I. 2002: Output costs of currency crisis and balance of payment crises in emerging markets. Comparative Economic Studies 44(2): 27-44.

Hutchison, M and Noy, I. 2005: How bad are twins? Output costs of currency and banking crises. Journal of Money, Credit and Banking 37(4): 725-752.

Hutchison, M, Noy, I and Wang, L. 2010: Fiscal and monetary policies and the cost of sudden stops. Journal of International Money and Finance 29(6): 973-987.

Jorda, O. 2005: Estimation and inference of impulse responses by local projections. American Economic Review 95(1): 161-182.

Judson, R and Owen, A. 1999: Estimating dynamic panel data models: A guide for macroeconomists. Economics Letters 65(1): 9-15.

Kamin, SB and Rogers, JH. 2000: Output and the real exchange rate in developing countries: An application to Mexico. Journal of Development Economics 61 (1): 85-109.

Kim, C, Morley, J and Piger, J. 2005: Nonlinearity and the permanent effects of recessions. Journal of Applies Econometrics 20(2): 291-309.

Krugman, P and Taylor, L. 1978: Contractionary effects of devaluation. Journal of International Economics 8(3): 445-456.

Lahiri, A and Vegh, CA. 2007: Output costs, currency crises, and interest rate defense of a peg. Economic Journal 117(516): 216-239.

Laeven, LA and Valencia, F. 2008: Systematic banking crises: A new database. IMF Working Paper No. 08/224. International Monetary Fund: Washington DC.

Luukkonen, R and Terasvirta, T. 1991: Testing linearity of economic time series against cyclical asymmetry. AnnaJes d'Economie et de Statistique 20(21): 125-142.

Morley, J and Piger, J. 2012: The asymmetric business cycle. The Review of Economics and Statistics 94(1): 208-221.

Morley, J. 1992: On the effect of devaluation during stabilization programs in LDCs. The Review of Economics and Statistics 74(1): 21-27.

McKinnon, R and Pill, H. 1995: Credible liberalizations and international capital flows: The 'over-borrowing syndrome'. In: Ito, T and Krueger, AO (eds). Financial deregulation and integration in East Asia. Chapter 1. University of Chicago Press: Chicago.

Neftci, SN. 1984: Are economics time-series asymmetric over the business cycles. Journal of Political Economy 92(2): 307-28.

Nickell, S. 1981: Biases in dynamic models with fixed effects. Econometrica 46(6): 1417-26.

Park, C and Lee, JW. 2001: Recovery and sustainability in East Asia. NBER Working Paper No. 8373. National Bureau of Economic Research: Cambridge, MA.

Sichel, DE. 1994: Inventories and the three phases of business cycle. Journal of Business and Economic Statistics 12(3): 269-77.

Teulings, C and Zabanov, N. 2014: Is economic recovery a myth? Robust estimation of impulse responses. Journal of Applied Econometrics 29(3): 497-514.

Velasco, A. 1987: Financial and balance of payments crisis: A simple model of the southern cone experience. Journal of Development Economics 27(1-2): 263-83.

Vuletin, G and Ruiz Pozuelo, J. 2014: Brookings, 4 February, http://www.brookings.edu/blogs/upfront/posts/2014/02/04argentina-currency-devaluation-vuletin, accessed 28 March.

World Bank. 2014: World Development?Indicators Online (WD1) database, http://data.worldbank. org/data-catalog/world-development-indicators, accessed 28 August.

SHEIDA TEIMOURI & TAGGERT J BROOKS

Department of Economics, College of Business Administration, University of Wisconsin-La Crosse, 1725 State Street, La Crosse, WI 54601, USA.

E-mails: steimouri@uwlax.edu; tbrooks@uwlax.edu

(1) Argentina intervened in the foreign exchange market by selling approximately $2.3 billion of its foreign exchange reserves. However, on 23 January, the country eventually experienced the largest devaluation since the crisis of 2001 (Vuletin and Ruiz Pozuelo, 2014). Turkey and South Africa raised the short-term interest rate to defend their currencies.

(2) To our knowledge, Erler et al. (2014) is the only empirical paper that separately evaluates the output costs associated with currency crashes, successful defenses, and unsuccessful defenses. However, they focus on a sample of 32 emerging countries and use a different methodology. The main question of their paper is to identify the least costly strategy a central bank can take in face of a speculative attack.

(3) Central banks' interventions in the foreign exchange market are often sterilized to avoid a decrease in the monetary base. But sterilized interventions do not eliminate the original cause of a speculative attack on the exchange rate. Therefore, without a reduction in the monetary base and an increase in domestic interest rates, sterilized interventions lead to further losses of foreign exchange reserves. Under severe speculative pressure, the only way to limit the capital flight is through unsterilized interventions, which are costly in terms of growth (Lahari and Vegh 2007).

(4) See Angkinand et al. (2006) for a survey on measures of currency crises.

(5) The threshold is usually the pooled (or country specific) mean of EMPI plus two (or three) standard deviations of EMPI.

(6) We exclude the interest rate component of EMPI due to lack of reliable interest rate data for a large part of our sample.

(7) They use the percentage change of end-of-period official nominal bilateral dollar exchange rate.

(8) It is possible that EMPI signals a crisis in case of a moderate depreciation and a small reserve loss. Therefore, in order to make this dummy variable more comparable with EVENT, we add another restriction: the depreciation must be at least 15%. The results are similar when we exclude this restriction.

(9) The CRISIS dummy variable is only equal to one if the depreciation component of EMPI is at least 75% of the index. If depreciation is large but not at least 75% of EMPI, then CRISIS dummy is equal to zero.

(10) In order to make sure these episodes are not episodes of large depreciations accompanied by reserve losses, we add another restriction to our identification of reserve crisis: the exchange rate depreciation in these episodes must also be less than 15%. The results are similar when we exclude this restriction or change the threshold to different levels (10% and 5%).

(11) Including a control group of countries that have never experienced a currency crisis does not change any of our results.

(12) Nickell (1981) points out that fixed effect estimation is biased when the lagged dependent variable is included in the model. However, the order of bias is reversely related to the time dimension (7), and thus not serious when T is large. In our case, similar to Cerra and Saxena (2008), T is large (53 years), and thus the bias is negligible. For example, for a panel of size N=100, T=30, and low persistence, Judson and Owen (1999) report that the bias of the least squares dummy variable (LSDV) estimator is approximately 2%-3% on the lagged dependent variable and less than 1% on other regressors.

(13) The results are not reported for brevity but are available in Table A1 in the Appendix.

(14) This is similar to Basistha and Teimouri (2015) who investigate the long-run cost associated with currency crashes. They find that currency crashes that are not accompanied by a banking crisis are expansionary in the long run.

(15) Higher interest rates increase the cost of borrowing and engaging in the currency carry-trade.

(16) We exclude all episodes of reserves crises that were accompanied or proceeded by a devaluation of 15% or larger. The results remain similar when we change the devaluation threshold to 10% or 5%.

(17) Teulings and Zabanov (2014) note that the analytical IRFs are especially sensitive to the number of lags of the dummy variable included in the ARDL model.

(18) Our estimation method follows Furceri and Zdzienicka (2012a, b); however, we estimate the immediate response of output to a currency crisis as well.

Table 1(a): Data description

Data       Definition                                   Count

LV         A nominal exchange rate depreciation of        183 (170) (a)
           30% or more which is also 10% higher than
           the rate of depredation in the previous
           period with a 5-year window around each
           crisis (Laeven and Valenda, 2008).

EVENT      The same as 1.1/with no window around each     317 (287)
           crisis. EMPI=Tbe percentage depreciation
           in the exchange rate + the percentage loss
           in foreign exchange reserves.

CS         CS = 1 if EMPI > Upper quartile of all       1,303 (1231)
           observations CS = 0 if EMPI < Upper
           quartile of all observations in the panel
           (Cerra and Saxena, 2008).

CRISIS     CRISIS = 1 if CS=l & dle/EMPI > 75 % & dle     306 (297)
           > 15 CRISIS = 0 Otherwise

CRISISR    CRISISR = 1 if C5 = l & dlr/EMPI > 75 % &      618 (567)
           dle < 15 CRISISR = 0 Otherwise (b)

CRISISR2   CRISISR2 = 0 if [dle.sub.t+1] > 15 or          468 (419)
           [dle.sub.t+2] > 15 CRISISR2 = CRISISR
           Otherwise

(a) All the numbers in parenthesis represent number of crises
for which GDP growth data is available.

(b) Where die represents percentage change in the exchange
rate and dir represent the percentage loss in foreign exchange
reserves. EMPI is the 'Exchange Market Pressure Index' and
represents the average of dlr and dle.

Table 1(b): Comparison of currency crises dummies

                    CRISIS

                         0         1
EVENT          0     3,884       151
               1       122       155

CRISISR        0     3,388       306
               1       618         0

Table 2: Models estimates using EVENT dummy

Dependent variable:         (1)              (2)
GDP growth                L-shaped         U-shaped

[DELTA][y.sub.i,t-1]        0.214 ***        0.214 ***
                           (0.016)          (0.015)
[DELTA][y.sub.i,t-2]        0.069 ***        0.073 ***
                           (0.016)          (0.016)
[DELTA][y.sub.i,t-3]        0.049 ***        0.054 ***
                           (0.016)          (0.016)
[DELTA][y.sub.i,t-4]       -0.002            0.004
                           (0.016)          (0.016)
[EVENT.sub.t]              -2.214 ***       -2.301 ***
  ([[gamma].sub.1])        (0.423)          (0.423)
[lambda]                                    -0.400 ***
                                            (0.086)
[EVENT.sub.t_1]

[EVENT.sub.t_2]

[EVENT.sub.t_3]

[EVENT.sub.t_4]

Constant                    3.360 ***        2.225 ***
                           (0.841)          (0.840)
SIC                    26,966.4         26,952.3
AIC                    26,625.4         26,605.1
N                       4,079            4,079
[R.sup.2]                   0.134            0.138

Dependent variable:         (3)              (4)
GDP growth                V-shaped        Cerra and
                                            Saxena
[DELTA][y.sub.i,t-1]        0.213 ***        0.209 ***
                           (0.015)          (0.016)
[DELTA][y.sub.i,t-2]        0.072 ***        0.070 ***
                           (0.016)          (0.016)
[DELTA][y.sub.i,t-3]        0.055 ***        0.055 ***
                           (0.016)          (0.017)
[DELTA][y.sub.i,t-4]        0.005            0.007
                           (0.016)          (0.016)
[EVENT.sub.t]              -1.623 ***       -2.296 ***
  ([[gamma].sub.1])        (0.440)          (0.423)
[lambda]                   -0.628 ***
                           (0.130)
[EVENT.sub.t_1]                              0.097
                                            (0.420)
[EVENT.sub.t_2]                              0.704 *
                                            (0.416)
[EVENT.sub.t_3]                              1.263 ***
                                            (0.414)
[EVENT.sub.t_4]                              1.594 ***
                                            (0.411)
Constant                    3.142 ***        3.176 ***
                           (0.840)          (0.839)
SIC                    26,950.6#        26,969.8
AIC                    26,603.4#        26,603.6
N                       4,079            4,079
[R.sup.2]                   0.139            0.140

Note: The dependent variable is the growth rate of GDP
([DELTA][y.sub.it]). The [[gamma].sub.1] represents the
coefficient of the contemporaneous crisis dummy (EVENT). [lambda]
* [[gamma].sub.1] represents the magnitude of bounce back in
output growth. Sample is based on the available data during the
1960-2013 period for 104 countries. Country and year fixed
effects are both included and the estimation method is two-way
panel fixed effects. AIC and SIC correspond to Akaike and Schwarz
criterion, respectively. The lower values of AIC and SIC
represent the preferred model (highlighted in bold). *, **, and
*** represent significant at 1%, 5%, and 10%, respectively. The
reported [R.sup.2] is within [R.sup.2].

The lower values of AIC and SIC represent the preferred
model indicated with #.

Table 3: Models estimates using EVENT and
controlling for debt and banking crises

Dependent variable:          (1)                (2)
GDP growth                 L-shaped           U-shaped

[DELTA][y.sub.i,t-1]        0.282 ***          0.281 ***
                           (0.016)            (0.016)
[DELTA][y.sub.i,t-2]        0.026              0.030 *
                           (0.016)            (0.016)
[DELTA][y.sub.i,t-3]        0.045 ***          0.050 ***
                           (0.016)            (0.016)
[DELTA][y.sub.i,t-4]       -0.006              0.001
                           (0.015)            (0.015)
[EVENT.sub.t]              -1.808 ***         -1.892 ***
  ([[gamma].sub.1])        (0.414)            (0.413)
Banking crisis             -1.625 ***         -1.705 ***
                           (0.422)            (0.421)
Debt crisis                -2.373 **          -2.169 **
                           (0.926)            (0.924)
[lambda]                                      -0.499 ***
                                              (0.102)
[EVENT.sub.t_1]

[EVENT.sub.t_2]

[EVENT.sub.t_3]

[EVENT.sub.t_4]

Constant                    2.318 ***          3.076 ***
                           (0.808)            (0.807)
SIC                    25,304.3           25,287.5
AIC                    24,966.2           24,943.1
N                       3,873              3,873
[R.sup.2]                   0.172              0.177

Dependent variable:          (3)                (4)
GDP growth                 V-shaped       Cerra and Saxena

[DELTA][y.sub.i,t-1]        0.280 ***          0.278 ***
                           (0.016)            (0.016)
[DELTA][y.sub.i,t-2]        0.029 *            0.028 *
                           (0.016)            (0.016)
[DELTA][y.sub.i,t-3]        0.050 ***          0.050 ***
                           (0.016)            (0.016)
[DELTA][y.sub.i,t-4]        0.001              0.003
                           (0.015)            (0.015)
[EVENT.sub.t]              -1.165 ***         -1.891 ***
  ([[gamma].sub.1])        (0.431)            (0.414)
Banking crisis             -1.730 ***         -1.665 ***
                           (0.421)            (0.421)
Debt crisis                -2.306 **          -2.195 **
                           (0.923)            (0.924)
[lambda]                   -0.922 ***
                           (0.176)
[EVENT.sub.t_1]                                0.278
                                              (0.407)
[EVENT.sub.t_2]                                0.717 *
                                              (0.402)
[EVENT.sub.t_3]                                1.201 ***
                                              (0.400)
[EVENT.sub.t_4]                                1.548 ***
                                              (0.398)
Constant                    3.036 ***          3.077 ***
                           (0.807)            (0.807)
SIC                    25,284.0#          25,306.6
AIC                    24,939.6#          24,943.4
N                       3,873              3,873
[R.sup.2]                   0.178              0.178

Note: The dependent variable is the growth rate of GDP
([DELTA][y.sub.it]). The [[gamma].sub.1] represents the coefficient
of the contemporaneous crisis dummy (EVENT). [lambda] *
[[gamma].sub.1] represents the magnitude of bounce back in output
growth. Banking crisis and Debt crisis are dummies, which are equal
to one if there is a banking or debt crisis in a country in a year.
Sample includes 104 countries during the 1960-2013 period. Country
and year fixed effects are both included and the estimation method
is two-way panel fixed effects. AIC and SIC correspond to Akaike
and Schwarz criterion, respectively. The lower values of AIC and
SIC represent the preferred model (highlighted in bold). *, **, and
*** represent significant at 1%, 5%, and 10%, respectively. The
reported [R.sup.2] is within [R.sup.2].

Note: The lower values of AIC and SIC represent the preferred
model (highlighted in bold) indicated with #.

Table 4: Model estimates using CRISIS dummy

Dependent variable:         (1)              (2)
GDP growth                L-shaped         U-shaped

[DELTA][y.sub.i,t-1]        0.141 ***        0.142 ***
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-2]        0.051 ***        0.052 ***
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-3]        0.022            0.024
                           (0.018)          (0.018)
[DELTA][y.sub.i,t-4]        0.003            0.007
                           (0.017)          (0.017)
[CRISIS.sub.t]             -2.103 ***       -2.310 ***
  ([[gamma].sub.1])        (0.435)          (0.442)
[lambda]                                    -0.202 **
                                            (0.079)
[CRISIS.sub.t_1]

[CRISIS.sub.t_2]

[CRISIS.sub.t_3]

[CRISIS.sub.t_4]

Constant                    2.961 ***        2.853 ***
                           (1.037)          (1.037)
SIC                    23,542.8         23,544.3
AIC                    23,208.6         23,203.9
N                       3,602            3,602
[R.sup.2]                   0.086            0.088

Dependent variable:         (3)               (4)
GDP growth                V-shaped      Cerra and Saxena

[DELTA][y.sub.i,t-1]        0.141 ***        0.140 ***
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-2]        0.051 ***        0.052 ***
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-3]        0.024            0.025
                           (0.018)          (0.018)
[DELTA][y.sub.i,t-4]        0.007            0.007
                           (0.017)          (0.017)
[CRISIS.sub.t]             -1.702' **       -2.259 ***
  ([[gamma].sub.1])        (0.454)          (0.445)
[lambda]                   -0.371 ***
                           (0.124)
[CRISIS.sub.t_1]                             0.017
                                            (0.446)
[CRISIS.sub.t_2]                             0.633
                                            (0.441)
[CRISIS.sub.t_3]                             0.629
                                            (0.434)
[CRISIS.sub.t_4]                             0.539
                                            (0.425)
Constant                    2.800 ***        2.846 ***
                           (1.037)          (1.037)
SIC                    23,541.6#        23,567.6
AIC                    23,201.2#        23,208.6
N                       3,602            3,602
[R.sup.2]                   0.089            0.088

Note: The dependent

Note: The dependent variable is the growth rate of GDP
([DELTA][y.sub.it]). The represents the coefficient of the
contemporaneous crisis dummy (CRISIS) and [lambda] *
[[gamma].sub.1] represents the magnitude of bounce back in output
growth. Sample includes 104 countries during the 1960-2013 period.
Country and year fixed effects are both included and the estimation
method is two-way panel fixed effects. AIC and SIC correspond to
Akaike and Schwarz criterion, respectively. The lower values of AIC
and SIC represent the preferred model (highlighted in bold). *, **,
and *** represent significant at 1%, 5%, and 10%, respectively. The
reported [R.sup.2] is within [R.sup.2].

Note: The lower values of AIC and SIC represent the preferred model
(highlighted in bold) indicated with #.

Table 5: Model estimates using CRISISR dummy

Dependent variable:         (1)              (2)
GDP growth                L-shaped         U-shaped

[DELTA][y.sub.i,t-1]        0.145 ***        0.138 ***
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-2]        0.054 ***        0.050 ***
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-3]        0.024            0.021
                           (0.018)          (0.018)
[DELTA][y.sub.i,t-4]        0.005            0.005
                           (0.017)          (0.017)
[CRISIS.sub.t]             -0.856 ***       -0.884 ***
  ([[gamma].sub.1])        (0.316)          (0.315)
[lambda]                                     0.844 ***
                                            (0.179)
[CRISIS.sub.t_1]

[CRISIS.sub.t_2]

[CRISIS.sub.t_3]

[CRISIS.sub.t_4]

                            3.061 ***        3.691 ***
Constant                   (1.041)          (1.046)
SIC                    23,559.5         23,544.4#
AIC                    23,225.3         23,204.0
N                       3,602            3,602
[R.sup.2]                   0.082            0.088

Dependent variable:         (3)               (4)
GDP growth                V-shaped      Cerra and Saxena

[DELTA][y.sub.i,t-1]        0.140 ***        0.139 ***
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-2]        0.052 ***        0.053 ***
                           (0.017)          (0.017)
[DELTA][y.sub.i,t-3]        0.021            0.021
                           (0.018)          (0.018)
[DELTA][y.sub.i,t-4]        0.007            0.004
                           (0.017)          (0.017)
[CRISIS.sub.t]             -1.400 ***       -0.857 ***
  ([[gamma].sub.1])        (0.346)          (0.315)
[lambda]                    0.474 ***
                           (0.123)
[CRISIS.sub.t_1]                            -1.573 ***
                                            (0.313)
[CRISIS.sub.t_2]                            -0.195
                                            (0.314)
[CRISIS.sub.t_3]                            -0.466
                                            (0.313)
[CRISIS.sub.t_4]                            -0.764 **
                                            (0.310)
                            3.666 ***        4.672 ***
Constant                   (1.051)          (1.050)
SIC                    23,552.2         23,557.6
AIC                    23,211.8         23,198.7#
N                       3,602            3,602
[R.sup.2]                   0.086            0.091

Note: The dependent variable is the growth rate of GDP
([DELTA][y.sub.it]). The [[gamma].sub.1] represents the coefficient
of the contemporaneous crisis dummy (CRISISR) and [lambda] *
[[gamma].sub.1] represents the magnitude of bounce back or drag in
output growth. Sample is based on available data during the
1960-2013 period for 104 countries. Country and year fixed effects
are both included and the estimation method is two-way panel fixed
effects. AIC and SIC correspond to Akaike and Schwarz criterion,
respectively. The lower values of AIC and SIC represent the
preferred model (highlighted in bold). *, **, and *** represent
significant at 1%, 5%, and 10%, respectively. The reported
[R.sup.2] is within [R.sup.2].

Note: The lower values of AIC and SIC represent the preferred model
(highlighted in bold) indicated with #.
COPYRIGHT 2015 Association for Comparative Economic Studies
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2015 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Teimouri, Sheida; Brooks, Taggert J.
Publication:Comparative Economic Studies
Article Type:Report
Geographic Code:1USA
Date:Mar 1, 2015
Words:10782
Previous Article:Reestimating China's underestimated consumption.
Next Article:Do indirect questions reduce lying about corruption? Evidence from a quasi-field experiment.
Topics:

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters