Information about Asia and the Pacific Asia y el Pacífico
Journal Issue
Share
Article

Did Output Recover from the Asian Crisis?

Author(s):
International Monetary Fund. Research Dept.
Published Date:
April 2005
Share
  • ShareShare
Information about Asia and the Pacific Asia y el Pacífico
Show Summary Details

The Asian financial crisis of 1997 generated a plethora of research that analyzed the causes of the crisis, but less attention has been paid to the aftermath. How long do crises last and to what extent does output recover? Although there is copious evidence that a financial crisis induces a recession, the literature has not examined whether a recession following a crisis permanently lowers the level of output. This paper analyzes whether the output reduction after the Asian crisis was a temporary deviation downward from the trend level that was eventually reversed as output reverted to trend (that is, the recession temporarily lowered the level of output) or, alternatively, the level of output shifted down permanently.

The paper approaches the question using a regime-switching common factor model. Recessions are decomposed into permanent and temporary components in a multivariate framework by introducing different state variables that control recoveries and recessions for each of the two components. Asymmetric adjustment in the temporary component is allowed to model the temporary “pluck” down from trend. We discuss these concepts in the context of their origin in the U.S. business cycle literature and present the data and results.

I. Theory and Literature Review

The causes of the Asian crisis have been fairly extensively discussed in the literature. For example, Corsetti, Pesenti, and Roubini (1998); Kochhar, Loungani, and Stone (1998); Radelet and Sachs (1998); and Berg (1999) provide overviews of the origins, onset, and spread of the crisis. This literature points to several factors that contributed to the crisis. Poor financial sector supervision and weak prudential regulation facilitated excessive lending, much of it directed toward real estate, construction, stock purchase, and consumer loans. The prolonged maintenance of pegged exchange rates encouraged foreign-currency-denominated liabilities. As the crisis approached, the ratio of short-term debt to foreign exchange reserves rose to high levels. When investors lost confidence in the economy and the currency, stock market values fell and exchange rates depreciated sharply. Interest rates spiked, reflecting the rise in risk premia. These developments led to bankruptcies among banks and finance companies as loans soured.

The behavior of recessions and subsequent recoveries from economic crises has not been studied as extensively as the causes. Some exceptions include cross-country studies by Barro (2001) and Park and Lee (2002). Barro does not detect a persistent adverse influence of currency and banking crises on long-run economic growth; Park and Lee find that a V-shaped pattern of growth is associated with crises. The countries hit by the Asian crisis experienced recessions of varying intensities. Output and consumption declined, and investment was hit especially hard. This study examines whether the Asian crisis had a long-run impact on the level, rather than the growth rate, of output (see Figure 1).

Figure 1.Actual Output, Consumption, and Investment

(Log level)

Note: The x-axis data labels refer to year:quarter.

In contrast to the general scarcity of studies on the behavior of crisis-driven recessions and recoveries, a considerable amount of research has been devoted to examining the properties of business cycles in the United States. Much of the literature focuses on (1) incorporating the idea of co-movement across economic time series, using the dynamic linear factor models created by Stock and Watson (1989, 1991, 1993) and (2) probing the idea of asymmetry, using the regime-switching approach pioneered by Hamilton (1989).

Regime switching has spurred a considerable debate on the nature of U.S. business cycle fluctuations. Two general types of parametric time-series models have been proposed, which have vastly different implications for the welfare effects of recessions (see Figure 2).

Figure 2.Output in Hamilton and Friedman Recessions

In Hamilton’s model (1989), the stochastic trend in output undergoes regime switching between positive and negative growth states. Because the regime switch occurs in the growth rate of the permanent component, a negative state results in a permanent output loss.

The second model assumes that regime switching occurs in a common temporary component. This idea has its roots in the work of Friedman (1964, 1993), in which a recession can be characterized as a temporary “pluck” down of output. After this large negative transitory shock dissipates, output returns to trend in a high-growth recovery phase. Because this type of recession represents a temporary deviation from trend followed by a full recovery to trend, the output loss is temporary.

The analysis in this paper draws on these concepts and on debates about the U.S. business cycle. The crisis-induced recessions in the Asian countries involved a simultaneous decline in several economic variables, which motivates the use of a common factor model like those used in the business cycle literature. In this paper, we are primarily interested in studying whether the asymmetry between expansions and crisis-driven recessions is more consistent with the Hamilton or the Friedman model. Both models involve V-shaped growth recoveries, although the Friedman model suggests that growth would be temporarily higher during recovery than during a normal expansion. This paper explores whether output springs back up to its original path following a crisis-driven recession or growth simply recovers to its trend rate, implying that the level of output has been permanently reduced compared with the original path. If the crisis leads to a temporary disruption in economic conditions or a temporary reduction in capacity utilization or employment, output could revert to its original path. In fact, if the crisis induces beneficial reforms, output may even recover to a higher path than before the crisis. In contrast, a switch to a lower state in the permanent component would imply a permanent output loss and could be characteristic, for example, of a reduction in productivity. After the crisis, productivity growth would resume, but there would be a permanent wedge in the level compared with pre-crisis forecasts.

A variety of methods and model specifications have been used to study the nature of the U.S. business cycle. Studies have been conducted using various assumptions about the source of asymmetry and with varying numbers of states. However, because of the use of univariate analysis, most of the literature that investigates asymmetry considers regime switching in only a temporary or a permanent component. Two exceptions are Kim and Murray (2002) and Kim and Piger (2002), which investigate the co-movement of several economic series and asymmetry in both temporary and permanent common factors. Kim and Piger, although specifying that output contains both permanent and temporary components, use only one state variable to control both components. Consequently, each recession is constrained to incorporate both temporary and permanent explanations. As in Kim and Murray, this paper uses a model that has two independent state variables for the temporary and permanent factors. This model is considered to be superior to a single-state-variable model with two or three states, as it allows researchers to determine whether a recession involves regime switches in the temporary or permanent components of output. However, Kim and Murray use a series of variables intended to capture co-movement with industrial production and focus on constructing a coincident indicator. As in Kim and Piger, this paper uses output, investment, and consumption, which theory predicts should share a common stochastic trend.

II. Econometric Model

In this section we present the specification of the dynamic two-factor model used for the empirical analysis. The logs of each of the three series of interest can be decomposed into a deterministic component, DTi, a permanent component, Pit, and a transitory component, Tit.

Yit=DTi+Pit+TitDTi=ai+DitPit=γint+ζitTit=λixt+ωit,

where Y=[output,investment,consumption],n is the common permanent component, x is the common temporary component, and ζ and ω are the independent idiosyncratic permanent and temporary components, respectively. The model can be written in differenced deviations from means as follows:

Δyit=γiΔnt+λiΔxt+zit,

where zit = Δζit + Δωit is a stationary composite of the idiosyncratic components, and γi, and λi are the factor loadings on the common permanent and common transitory components, respectively, for i = [output, investment, and consumption].

The growth rate of the common permanent component is stationary and is approximated by a second-order autoregressive process, AR(2). Note that a stationary growth rate implies that the level is nonstationary, in accordance with the definition of a stochastic trend. In addition, there is a trend, β, that depends on the permanent state, S1t:

Δnt=βS1t+φ1Δnt1+φ2Δnt2+vt,vti.i.d.N(0,1).

The state-dependent trend introduces asymmetry along the lines of Hamilton (1989):

βS1t=β0+β1S1t;S1t={0,1}.

During an expansion phase (S1t = 0), the stochastic trend grows with the drift rate β0. If β1 is negative, the trend shifts to a lower growth state when S1t = 1 and shifts to a recession phase if β0 + β1 < 0.

The common temporary component is stationary in its levels and is approximated by a second-order autoregressive process. To incorporate Friedman’s type of asymmetry, we allow the temporary component to undergo regime switching in response to a second state variable, S2t:

xt=τS2t+φ11xt1+φ12xt2+ut,uti.i.d.N(0,1).

In state S2t = 0, the intercept is zero. If τi, < 0, the economic series is plucked down when S2t = 1. When the state returns to normal, S2t = 0, the economy reverts back to trend level.

Finally, each series has its own stationary idiosyncratic component, again approximated by an AR(2).1

zit=Ψi1zit1+Ψi2zit2+eit,eiti.i.d.N(0,1)E(vruseit)=0,i,r,s,t.

Both state variables are assumed to be independent first-order Markov switching processes with transition probabilities given by

P[S1t=0|S1t1=0]=q1,P[S1t=1|S1t1=1]=p1,andP[S2t=0|S2t1=0]=q2,P[S2t=1|S2t1=1]=p2.

III. Econometric Analysis and Results

We use the available quarterly data from six Asian countries (Hong Kong SAR, Indonesia, Korea, the Philippines, Malaysia, and Singapore).2 In particular, we take the logs of GDP, gross fixed capital formation, and private consumption in constant prices, and seasonally adjust them using Census X-12. (The data sources are described in Appendix I.) The number of available time-series observations ranges from 32 quarters to 89 quarters. However, stacking the three related economic variables in a common factor model effectively triples the number of observations.3

Augmented Dickey-Fuller and Phillips-Perron tests provide strong evidence that each of these series contains a unit root (see Table 1). Indeed, the null hypothesis of a unit root cannot be rejected for any of the variables in levels at the 5 percent significance level and can be rejected only at the 10 percent level for Hong Kong SAR’s private consumption. The unit root hypothesis can easily be rejected at the 1 percent level for all variables in changes.

Table 1.Unit Root Tests 1
CountryObs.VariableADF StatADFP-valPP StatPP P-val
Hong Kong SAR65LRGDP−2.050.26−2.600.10
LR1NV−2.050.26−2.070.26
LRCON−2.680.08−2.660.09
Indonesia32LRGDP−2.590.11−2.120.24
LR1NV−2.310.18−1.630.46
LRCON1.800.37−1.800.38
Korea89LRGDP−0.970.76−0.950.77
LRINV−1.170.68−1.090.72
LRCON−0.610.86−0.660.85
Malaysia41LRGDP−2.430.14−1.940.31
LRINV−1.980.29−2.030.27
LRCON−1.100.71−1.130.69
Philippines85LRGDP1.011.000.450.98
LRINV−1.380.59−1.260.65
LRCON0.951.000.971.00
Singapore89LRGDP−0.910.78−1.000.75
LRINV−1.380.59−1.390.59
LRCON−0.810.81−0.790.82

Variables are defined in Appendix I. The lag lengths for the Augmented Dickey-Fuller tests were selected on the basis of the Schwartz Information Criterion, and the bandwidth for the Phillips-Perron tests were based on the Newey-West method using the Bartlett kernel. Critical values are from MacKinnon. The results shown are based on unit root tests in levels, with a constant. All series were stationary in first differences.

Variables are defined in Appendix I. The lag lengths for the Augmented Dickey-Fuller tests were selected on the basis of the Schwartz Information Criterion, and the bandwidth for the Phillips-Perron tests were based on the Newey-West method using the Bartlett kernel. Critical values are from MacKinnon. The results shown are based on unit root tests in levels, with a constant. All series were stationary in first differences.

Standard theoretical models of capital accumulation in an intertemporal optimizing framework imply that output, investment, and consumption share a common stochastic trend. The permanent income hypothesis would identify consumption with the stochastic trend. Indeed, some researchers (Kim and Piger, 2002) assume that consumption represents the stochastic trend in output. That restriction is not imposed here in order to allow for possible liquidity constraints that would make at least a fraction of the population consume out of current income and would thus imply a transitory component to consumption. The common temporary (or cyclical) component could reflect a variety of shocks, including those from supply- and demand-side sources.

The model outlined here can be written in state space form (Appendix II), which allows the application of a Kalman filter. The regime switch is estimated by Kim’s (1994) approximate maximum likelihood algorithm, which is a computationally efficient method of estimating Markov switching in both the observation and transition equation.

In Table 2, we show the maximum likelihood parameter estimates for the transition probabilities and state-dependent means, as well as the factor loadings on permanent and temporary components, and the autoregressive parameters and error standard deviations of the idiosyncratic components for output, investment, and consumption.4 All the factor loadings (γi) for output, investment, and consumption are positive on the permanent components for all the countries,5which suggests that the permanent component is well identified, with the three variables for each country exhibiting co-movement. The state-dependent mean on the permanent component is positive when S1t = 0 and negative when S1t = 1 (for example, β0 > 0 and (β0 + β1 < 0), identifying expansions and recessions. There is some evidence of binding liquidity constraints, as the factor loadings on the temporary components are greater for consumption than output for four of the six countries (and are statistically significant for Indonesia and Korea), indicating that consumption contains a cyclical fluctuation and, thus, that individuals are not fully capable of smoothing their consumption. The state-dependent mean (τ1) of the temporary component is negative except for Hong Kong SAR and Singapore, but the effect of a switch in S2t also depends on the sign of the factor loadings on the temporary components, which vary across countries and economic series.

Table 2.Maximum Likelihood Estimates
ParametersHong Kong SARIndonesiaKoreaMalaysiaPhilippinesSingapore
Transition probabilities:
q10.9820.9650.9890.9740.9660.953
(0.018)(0.034)(0.011)(0.026)(0.027)(0.026)
p10.4680.7290.0000.6500.7490.676
(0.412)(0.213)(0.001)(0.273)(0.143)(0.145)
q20.0020.2660.9470.8980.2960.783
(0.485)(0.231)(0.050)(0.090)(0.231)(0.075)
p20.4850.9010.8170.8610.8940.410
(0.645)(0.066)(0.128)(0.113)(0.086)(0.117)
AR(2) coefficients:
φ10.476−0.0260.185−0.064−0.1480.029
(0.166)(0.135)(0.069)(0.153)(0.099)(0.113)
φ20.0530.4900.062−0.001−0.0060.000
(0.156)(0.116)(0.065)(0.005)(0.007)(0.002)
φ110.7200.207−0.1790.1350.0830.473
(0.129)(0.123)(0.140)(0.132)(0.142)(0.075)
φ120.216−0.011−0.008−0.0050.0100.346
(0.130)(0.013)(0.012)(0.009)(0.900)(0.070)
ΨY10.3870.103−0.360−0.4701.6460.420
(0.460)(0.275)(0.120)(15.443)(0.403)(4.242)
ΨY20.136−0.003−0.032−0.055−0.677−0.044
(0.430)(0.014)(0.022)(3.650)(0.332)(0.890)
ΨI1−0.0670.0720.182−0.200−0.208−1.548
(10.348)(0.238)(0.166)(34.100)(32.351)(2.036)
ΨI20.0520.140−0.0080.022−0.010−0.599
(1.966)(0.232)(0.015)(2.416)(2.564)(1.591)
ΨC1−0.4390.8290.190−0.302−0.140−0.387
(0.153)(0.555)(2.050)(0.172)(0.180)(0.109)
ΨC2−0.048−0.172−0.0090.016−0.005−0.037
(0.033)(0.230)(0.203)(0.149)(0.013)(0.021)
Factor loadings on permanent components:
γY0.6010.2370.4160.6390.7300.673
(0.118)(0.131)(0.047)(0.073)(0.069)(0.057)
γI0.2150,2220.4150.5040.4630.313
(0.086)(0.131)(0.059)(0.063)(0.052)(0.060)
γC0.4470.1080.4660.4310.1760.380
(0.076)(0.060)(0.041)(0.083)(0.111)(0.057)
Factor loadings on temporary components:
λY0.0250.147−0.083−0.130−0.025−0.080
(0.088)(0.057)(0.049)(0.066)(0,083)(0.049)
λI0.8880.072−0.1180.3440.4220.486
(0.088)(0.056)(0.064)(0.095)(0.231)(0.067)
λC0.137−0.3510.1100.111−0.0870.029
(0.107)(0.085)(0.046)(0.089)(0.094)(0.060)
Standard errors:
σY0.3200.4570.5980.0000.0270.006
(0.171)(0.102)(0.058)(0.040)(0.046)(0.316)
σI0.0000.5990.5760.0000.0000.000
(0.015)(0.099)(0.081)(0.012)(0.324)(0.012)
σC0.6430.1880.0000.7180.9610.799
(0.070)(0.124)(0.013)(0.080)(0.080)(0.060)
State-dependent means:
β00.1210.9860.1630.3340.3960.406
(0.140)(0.636)(0.108)(0.175)(0.156)(0.137)
β1−4.170−8.458−14.300−4.770−3.417−3.520
(1.238)(4.976)(1.678)(1.042)(0.561)(0.599)
τ10.233−4.968−6.686−3.528−2.5062.984
(0.421)(1.248)(1.892)(0.988)(3.035)(0.446)
Log likelihood−107.176−39.662−98.418−56.094−153.867−157.621
Sample period1986:1–2002:11994:1–2001:41980:1–2002:11992:1–2002:11981:1–2002:11980:1–2002:1
Total observations19596267123255267
Note: Standard errors arc in parentheses.
Note: Standard errors arc in parentheses.

The expected duration of the expansion and contraction phases is shown in Table 3, as derived from the parameter estimates of the transition probabilities.6 For example, the expected durations for Hong Kong SAR are 57 quarters for the expansion phase of the permanent component and 2 quarters for the contraction phase of the permanent component. For all the countries, expansions are expected to last considerably longer than contractions. This finding is consistent with longstanding results in the U.S. business cycle literature. Indeed, Mitchell noted in 1927 that “business contraction seems to be a briefer and more violent process than business expansion.”

Table 3.Expected Durations of States Affecting the Permanent Component(Quarters)
Hong Kong SARIndonesiaKoreaMalaysiaPhilippinesSingapore
Expansion572987382921
Contraction241343

Figures 3 and 4 show the probabilities that the permanent and temporary common components, respectively, undergo a regime switch. It is evident from Figure 3 that the crisis induced a recession in the permanent components of all the countries at the time of the Asian crisis. The probability of being in the recession state reaches one in all the countries except the Philippines, for which it peaks at about 0.2. The Philippines instead endured a deep and prolonged recession in the early 1980s associated with the debt crisis and domestic turmoil. The recession state is short-lived in Korea but lasts for several quarters in Indonesia. Figure 5 illustrates the common permanent component for each of the countries.7

Figure 3.Probability of Permanent Recession

Note: The x-axis data labels refer to year:quarter.

Figure 4.Probability That Stat2t = 1

Note: The x-axis data labels refer to year:quarter.

Figure 5.Common Permanent Component

(Log level)

Note: The x-axis data labels refer to year:quarter.

The cumulative effects of regime switches in the permanent components of the six countries over the period 1997–99 are shown in Table 4.8 The magnitude of the losses from the Asian crisis is economically significant for all countries except perhaps the Philippines.

Table 4.Magnitude of Permanent Output Loss(Percent)
Hong Kong SARIndonesiaKoreaMalaysiaPhilippinesSingapore
Average permanent loss per recession quarter13.55.610.36.34.13.8
Cumulative loss in Asian crisis2 (excl. AR terms)7.022.310.319.01.512.9
Cumulative Loss in Asian Crisis (incl. AR terms)14.841.613.717.91.313.3

Difference in state-dependent mean, βS1t, when S1t, = 1 compared to S1t = 0, and adjusted for the factor-loading coefficient and normalization.

Reflects average loss multiplied by probability of permanent recession for the period 1997–99. Does not include the effects of the AR terms of the permanent component.

Difference in state-dependent mean, βS1t, when S1t, = 1 compared to S1t = 0, and adjusted for the factor-loading coefficient and normalization.

Reflects average loss multiplied by probability of permanent recession for the period 1997–99. Does not include the effects of the AR terms of the permanent component.

The parameter estimates shown in Table 2 indicate a lack of solid support for Friedman-style recessions with temporary output losses. More than half of the factor loadings on the temporary components are statistically insignificant, and the signs of the coefficients are inconsistent across the three economic series, except for Hong Kong SAR. However, where the magnitude and statistical significance of the λ coefficients are largest (consumption for Indonesia and Korea, and investment for Hong Kong SAR, Malaysia, and Singapore), the common temporary component declines sharply at the time of the crisis.9Figure 4 shows the probability that S2t = 1, which corresponds to a contraction (expansion for Indonesia) in consumption or investment, as just discussed. The probability that the recessions associated with the Asian crisis contained a temporary component is most evident for Indonesia, Korea, and Singapore, as shown in Figure 6.

Figure 6.Common Temporary Component

(Log level)

Note: The x-axis data labels refer to year:quarter.

The changes in actual output and the permanent component of output are shown in Figure 7 for each country. It is apparent that changes in the permanent component, including the effects of AR terms and deterministic drifts, account for most of the changes in actual output. Figure 8 isolates the contemporaneous effects of changes in the state-dependent mean (βS1t), excluding AR and deterministic drift terms. Clearly, the regime switch in the permanent component accounts for a considerable amount of the negative growth rate of output during the Asian crisis.

Figure 7.Changes in Output: Actual and Permanent Components

(Change in log level)

Note: The x-axis data labels refer to year:quarter.

Figure 8.Changes in Output: Actual and Component Attributed to State-Dependent Mean of Permanent Component1

(Change in log level)

Note: The x-axis data labels refer to year:quarter.

1 Contemporaneous effect only.

IV. Conclusions and Directions for Future Research

The chief objective of this paper has been to investigate whether output losses associated with the Asian crisis have been permanent or temporary. We used a two-common-factor model with regime switching in each of the factors and used real GDP, gross fixed capital formation, and private consumption to identify the common transitory and stochastic trends.

The results indicate some amount of permanent output loss in all countries, despite rapid returns to positive growth states. Output in most of the countries appears to behave according to Hamilton’s model, in which the growth rate of output switches between positive and negative growth states. The recovery phase is predominantly characterized by a return to the normal growth rate of an expansion rather than a higher-than-normal growth rate. Thus, the level of output is permanently lower than its initial trend path.

The nature of the output loss has various implications for the output gap and for policy response. A permanent loss is associated with a downward shift of potential output, whereas a temporary loss is associated with a deterioration of the output gap. Nevertheless, the appropriate policy response depends on the source of the loss and the effectiveness of macroeconomic and structural policies in stimulating potential output and reducing any distortions.

This paper is an attempt to understand the nature of recessions and recoveries from economic crises, but significant scope remains for further research, as follows:

  • • A wide range of methods has been used to examine the nature of U.S. business cycles; many of these methods could be brought to bear in studying crisis-driven recessions and recoveries. Our understanding of these recessions and recoveries could benefit from advances in the estimation and diagnostics of nonlinear models.
  • • This study is limited to the Asian crisis; many other episodes of financial crisis could be explored, with the research limited mainly by data availability.
  • • Future research could investigate the source of permanent loss in output, such as from a permanent rise in unemployment or decline in productivity. The banking crisis in Sweden in the early 1990s, for instance, appears to have induced a deep recession that involved a permanent increase in unemployment (Cerra and Saxena, 2000). Also, the source of productivity decline could be explored. For example, does productivity fall as a result of a collapse in financial intermediation that creates a wedge between savings and its efficient allocation?
  • • Future research could also explore the relationship between the frequency and magnitude of crises, and the relationship between the trend growth rate and the prevalence of crises. More relevant for policy analysis would be research on whether the magnitude of output loss and the behavior of the subsequent recovery are functions of economic policy responses and reforms.
APPENDIX I Data Sources
VariableCountrySampleSource
Real gross domestic product (RGDP)Hong Kong SAR (HK)

Indonesia (IDN)
1986:1–2002:1

1994:1–2001:4
WEFA

Buletin Statistik Bulanan (Monthly Statistical Bulletin), Indikator Ekonomi
Korea (KOR)

Malaysia (MYS)
1980:1–2002:1

1992:1–2002:1
Bank of Korea

Sharan Perangkann Bulanan (Monthly Statistical Abstract),

Department of Statistics
Philippines (PHL)

Singapore (SGP)
1981:1–2002:1

1980:1–2002:1
WEFA

WEFA
Real gross fixed capital formation (RINV)Hong Kong SAR (HK)

Indonesia (IDN)
1986:1–2002:1

1994:1–2001:4
WEFA

Buletin Statistik Bulanan (Monthly Statistical Bulletin), Indikator Ekonomi
Korea (KOR)

Malaysia (MYS)
1980:1–2002:1

1992:1–2002:1
Bank of Korea

Sharan Perangkann Bulanan

(Monthly Statistical Abstract),

Department of Statistics
Philippines (PHL)

Singapore (SGP)
1981:1–2002:1

1980:1–2002:1
WEFA

WEFA
Real personal consumption (RPCON)Hong Kong SAR (HK)

Indonesia (IDN)
1986:1–2002:1

1994:1–2002:1
WEFA

Buletin Statistical Bulanan

(Monthly Statistical Bulletin),

Indikator Ekonomi
Korea (KOR)

Malaysia (MYS)
1980:1–2002:1

1992:1–2002:1
Bank of Korea

Sharan Perangkann Buianan

(Monthly Statistical Abstract),

Department of Statistics
Philippines (PHL)

Singapore (SGP)
1981:1–2002:1

1980:1–2002:1
WEFA

WEFA
Note: WEFA = Wharton Econometric Forecasting Associates.
Note: WEFA = Wharton Econometric Forecasting Associates.
APPENDIX II State Space Representation

This section presents the state space representation of the model discussed in Section III.

Observation Equation:

[ΔytΔitΔct]=[γ10λ1λ1100000γ20λ2λ2001000γ30λ3λ3000010][ΔntΔnt1xtxt1z1tz1t1z2tz2t1z3tz3t1]

Transition Equation:

[ΔntΔnt1xtxt1z1tz1t1z2tz2t1z3tz3t1]=[β0+β1S1t0τS2t0000000]+[φ1φ200000000100000000000φ11φ1200000000100000000000Ψ11Ψ1200000000100000000000Ψ21Ψ2200000000100000000000Ψ31Ψ320000000010][Δnt1Δnt2xt1xt2z1t1z1t2z2t1z2t2z3t1z3t2]+[vt0ut0e1t0e2t0e3t0]

Covariance Matrix of the Disturbance Vector:

E([vt0ut0e1t0e2t0e3t0][vt0ut0e1t0e2t0e3t0])=[σv2000000000100000000000σu2000000000100000000000σe22000000000100000000000σe22000000000000000000000σe3200000000010]

REFERENCES

    BarroRobert J.2001Economic Growth in East Asia Before and After the Financial Crisis,NBER Working Paper No. W8330 (CambridgeMassachusetts: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    BergAndrew1999The Asia Crisis: Causes, Policy Responses and Outcomes,IMF Working Paper 99/138 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    CerraValerie and Sweta C.Saxena2000Alternative Methods of Estimating Potential Output and the Output Gap: An Application to Sweden,IMF Working Paper 00/59 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    CerraValerie and Sweta C.Saxena2002Contagion, Monsoons and Domestic Turmoil in Indonesia’s Currency Crisis,Review of International Economics Volume 10 Issue 1 pp. 3644.

    • Search Google Scholar
    • Export Citation

    CorsettiGiancarloPaoloPesenti and NourielRoubini1998What Caused the Asian Currency and Financial Crisis?NBER Working Paper No. 6833 (CambridgeMassachusetts: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    FriedmanMilton1964Monetary Studies of the National Bureau,The National Bureau Enters Its 45th Year 44th Annual Report 7-25 (New York: National Bureau of Economic Research); reprinted in Friedman Milton 1969 The Optimum Quantity of Money and Other Essays (Chicago: Aldine) pp. 26184.

    • Search Google Scholar
    • Export Citation

    FriedmanMilton1993The ’Plucking Model’ of Business Fluctuations Revisited,Economic Inquiry Vol. 31 (April) pp. 17177.

    • Search Google Scholar
    • Export Citation

    GarciaRené1998Asymptotic Null Distribution of the Likelihood Ratio Test in Markov Switching Models,International Economic Review Vol. 39 (August) pp. 76388.

    • Search Google Scholar
    • Export Citation

    HamiltonJames D.1989A New Approach to the Economic Analysis of Nonstationary Times Series and the Business Cycle,Econometrica Vol. 57 (March) pp. 35784.

    • Search Google Scholar
    • Export Citation

    HansenBruce E.1992The Likelihood Ratio Test Under Non-standard Conditions: Testing the Markov Switching Model of GNP,Journal of Applied Econometrics Vol. 7 pp. S61S82.

    • Crossref
    • Search Google Scholar
    • Export Citation

    KimChang-Jin1994Dynamic Linear Models with Markov Switching,Journal of Econometrics Vol. 60 (Jan.-Feb.) pp. 122.

    KimChang-Jin and Christian J.Murray2002Permanent and Transitory Components of Recessions,Empirical Economics Vol. 27 No. 2 pp. 16383.

    • Search Google Scholar
    • Export Citation

    KimChang-Jin and Charles R.Nelson1999State Space Models with Regime Switching: Classical and Gibbs-Sampling Approaches with Applications (CambridgeMassachusetts: MIT Press).

    • Export Citation

    KimChang-Jin and JeremyPiger2002Common Stochastic Trends, Common Cycles, and Asymmetry in Economic Fluctuations,Journal of Monetary Economics Vol. 49 (September) pp. 11891211.

    • Search Google Scholar
    • Export Citation

    KochharKalpanaPrakashLoungani and MarkStone1998The East Asian Crisis: Macro-economic Developments and Policy Lessons,IMF Working Paper 98/128 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    MitchellWesley C.1927Business Cycles: The Problem and Its Setting (New York: National Bureau of Economic Research).

    • Export Citation

    ParkYung Chul and Jong-WhaLee2002Recovery and Sustainability in East Asia,in Korean Crisis and Recoveryed. by David T.Coe and Se-JikKim (Washington: International Monetary Fund) pp. 35396.

    • Search Google Scholar
    • Export Citation

    RadeletSteven and JeffreySachs1998The Onset of the East Asian Financial Crisis,NBER Working Paper No. 6680 (CambridgeMassachusetts: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    StockJames H. and Mark W.Watson1989New Indexes of Coincident and Leading Indicators,in NBER Macroeconomics Annual 1989 Vol. 4ed. by Olivier J.Blanchard and S.Fischer (CambridgeMassachusetts: MIT Press) pp. 35193.

    • Search Google Scholar
    • Export Citation

    StockJames H. and Mark W.Watson1991A Probability Model of the Coincident Economic Indicators,in Leading Economic Indicators: New Approaches and Forecasting Recordsed. by KajalLahiri and Geoffrey H.Moore (New York: Cambridge University Press) pp. 6385.

    • Search Google Scholar
    • Export Citation

    StockJames H. and Mark W.Watson1993A Procedure for Predicting Recessions with Leading Indicators: Econometric Issues and Recent Experiences,in Business Cycles Indicators and Forecastinged. by James H.Stock and Mark W.Watson (Chicago: University of Chicago Press) pp. 95156.

    • Search Google Scholar
    • Export Citation
*Senior Economist, IMF Institute; and Assistant Professor, Graduate School of Public and International Affairs, University of Pittsburgh, respectively. The authors gratefully acknowledge Chang-Jin Kim, Chris Murray, and Jeremy Piger for providing original Gauss programs upon which the program used in this paper was based. They would also like to thank Bas Bakker, Craig Beaumont, Peter Berezin, Tarhan Feyzioglu, Munehisa Kasuya, Kalpana Kochhar, Papa N’Diaye, seminar participants at the IMF Institute and the Asian Crisis III conference in Tokyo, Japan, and an anonymous referee for helpful suggestions.
1The assumption of unitary variance is made for identification, but the assumption is not particularly restrictive, as the variances of the permanent and temporary components of output, investment, and consumption depend on the magnitude of the factor loadings.
2Quarterly data for Thailand were available for only three years, so it was dropped from consideration.
3See Kim and Nelson (1999), Chapter 3, for an application of the dynamic common factor model to four coincident indicators in U.S. data.
4The common factor model makes use of the stacked information from the vector of variables: out put, investment, and consumption. For each of the countries shown in Table 4, the complete set of factor loadings, and idiosyncratic AR parameters and error standard deviations are presented for the vector of variables. Several sets of initial values were employed to ensure the robustness of the results.
5Testing for the number of states in Markov switching models is complicated by a number of problems, particularly, nuisance parameters under the null hypothesis and a singular Hessian. If the nuisance parameters exist only under the alternative hypothesis but not under the null hypothesis, the likelihood ratio, LM, and Wald tests cannot be applied. In this particular model, some of the AR parameters and transition probabilities are unidentified under the null hypothesis that all of the gammas are zero, or that all of the lambdas are zero. Hansen (1992) and Garcia (1998), among others, have considered the problem of nuisance parameters under the null, but the distribution of the test statistic for the state space model employed in this paper is unknown when nuisance parameters exist only under the alternative hypothesis. Nevertheless, there is scope for inference in the model: the hypothesis that any particular factor loading equals zero does not involve any unidentified parameters and standard distribution theory is valid. Moreover, while the estimations assume the existence of two state variables, there is no reason to pre-suppose the estimated permanent loss would be economically significant.
6Testing whether the transition probabilities, p and q, are zero or one, is complicated by the fact that if the parameter lies on the boundary, standard inference is invalid. As the expected duration of a state becomes either long-lasting or of very short duration, the associated transition probability would lie close to a boundary value.
7The prevalence of permanent output losses in the Asian countries following their crises corroborates the findings that Sweden’s crisis in the early 1990s led to a large permanent output loss (Cerra and Saxena, 2000).
8These effects are the extent of contemporaneous output loss over 1997–99. To the extent that the sum of the AR coefficients on the permanent components are positive (negative), the output losses would continue to mount (would diminish) beyond the crisis period. The AR components (φ1 and φ2 in Table 2) sum to a positive number for all countries except Malaysia and the Philippines.
9The common temporary component increases for Indonesia, but λC is negative, thus the effect on consumption would be negative.

Other Resources Citing This Publication