Abstract
This paper contributes to the existing literature of currency crisis dating in a number of areas. Firstly, we combine the Monte Carlo simulation with a modified version of the Hill’s estimator to obtain robust results and deal with the bias–variance tradeoff in identifying extreme values. Secondly, to avoid sample- specific thresholds, we construct our results upon stationary series with the help of the Hill’s estimator. We also report the whole identified crisis episodes while disregarding the exclusion window technique, which may induce identification problems. To select the reference country when building the exchange market pressure, a statistical search between two exogenously given options is employed. Thirdly, different data frequencies are applied and the results are evaluated. Our findings suggest that higher frequency data are more appropriate when applying extreme value theory (EVT). Our results recommend researchers to be more cautious when applying EVT and interpreting tail incidences that are obtained from lower frequency data.
Similar content being viewed by others
Notes
Interestingly, governments deliberately do the realignments in tranquil periods to avoid future crises.
To adjust for instances where countries have high inflation, they also require that the depreciation be at least 10% higher than the previous year.
Interest rates can affect capital flows and speculative attacks.
Negative values show large appreciations or large increases in reserves that are considered fundamentally different from depreciation pressure crises.
Some researchers argue that transforming continuous variables into binary variables may result in loss of information. Thus, they treat EMP as a continuous dependent variable (Eliasson and Kreuter 2001).
As Eichengreen et al. mention the ideal weights should be the slope coefficients that reflect how much official intervention (change in international reserves and/or interest rate) would be required to avoid 1% change in the exchange rate. However, there is no reliable theoretical model for the foreign exchange that the profession agrees upon and the reduced form models provide a good fit.
Angkinand et al. (2006).
The weights can be driven from either each country-specific sample [own country precision weights; see \(e.g.\) Eichengreen et al. (1995) and Aziz et al. (2000)], or entire sample of countries [pooled precision weights; see \(e.g.\)Kaminsky and Reinhart (1998) and Glick and Hutchison (2001)]. Some researchers believe pooled weights may lessen this problem but at the cost of probable heterogeneity.
He also claims his method can overcome Flood and Marion’s (1999) argument. Following Krugman (1979), Flood and Marion argue since the interest rate falls back and reserves flow back right after the devaluation, these two effects may cancel out some part of changes in exchange rates and dampen the EMP index. So, in the case of predictable devaluations, the EMP index may fail to identify a crisis.
In economics and finance, the data come from Frechet domain of attraction that includes a range of distributions: Student’s \(t\), the stable distributions, and ARCH- type process.
During our sample period, the USD has experienced weak periods in the 1970s (1971–1973) and the mid-1990s.
For a comprehensive introduction and comparison of ARCH family models, see Enders (2004).
Case of Iceland can be a little confusing. While monthly data show Krona is marginally more stable in term of the USD, quarterly data indicate different direction and show that Krona is more stable in terms of the DM. Considering its major economic partners, our final selection is made based on quarterly data.
Each conditional variance of the three components is estimated based on the EGARCH(1,1) model: \(y_{t}=x_{t}{\varphi +\varepsilon }_{t}\) where ln \(\sigma _{t}^{2}=\nu _{0}+g(z_{t-1})+\nu _{1}ln\sigma _{t-1}^{2} ,{\varepsilon }_{t} =z_{t }\sigma _{t}, andg(z_{t})\) is a well-defined function of z\(_{t}\). In the mean equation, \(y_{t}\) represents one of the three components and \(x_{t }\) is a lagged values of \(y_{t}\). The fitted conditional standard deviation (\(\sigma ^{h}_{t})\) is used to generate weights in the EMP index. For conditional volatility, there is no concern about non-convergence.
Having correct size and high power results are two main desired factors in every statistical test. In unit root tests, the presence of cross-sectional correlation causes size distortion that leads to over-reject of the unit root null.
MSE of \(S\) simulated observations of the estimator X\(_{s}^{\sim }\) can be decomposed as MSE((X\(_{s}^{\sim })_{s=1,2, \ldots , S}\), X) = (X\(^{-}\) - X)\(^{2}\) + 1/\(S \quad \Sigma _{s=1}^{S}\)(X\(_{s}^{\sim }\) - X)\(^{2}\), where X\(^{- }\)represents the mean of \(S\) simulated observations. The first part measures the bias and second part the inefficiency.
The other class of distributions that can account for fat tails are stable laws. But as Wagner and Marsh (2005) argue, although symmetric stable laws with \(\alpha <\)2 are theoretically justified for extreme value theory, applications of Hill’s estimator do not appear promising for stable laws in small samples.
To identify the stable region of the estimated modified Hill’s indices, we compute \(\gamma (m)\) for a range of values of \(m\) from 1 to M, where \(M\le n/2\). The region where \(\gamma \) varies almost linearly in \(m\) will be defined as the stable region.
With the help of the identified stable region, the number of Monte Carlo simulations can be potentially reduced. One can run the Monte Carlo only within the stable region and avoid the massive simulations. However, in practice, our main concern was to obtain more reliable results (with the highest possible degree of scrutiny) rather than decreasing the number of Monte Carlo simulations. Thus, we run the simulation for all \(\alpha \) from 1 to 10 (with increment of 0.1) and check the entire results. But at the final stage, we picked the most significant optimum cutoff value (\(m\)**) merely from the region of stable modified Hill’s estimations. If it had been time consuming and difficult to run we might have narrowed our simulation scope only to the stable region.
In the recursive method, first, each series is arranged in descending order and the Hill’s estimator are computed for the first thirtieth percentile. In the next step, the computed amounts for Hill’s index are regressed on a constant and time trend variable successively. Then, the recursive residuals are derived to find the structural break. The optimum \(m \)is picked where the value of recursive residuals lies outside of the two-standard errors band.
In this paper, the quarters of crisis episodes are identified based on monthly and quarterly data. If annual data is of interest, Bordo et al. (2001) is an excellent source of comparison.
References
Abiad A (2003) Early-warning systems: a survey and a regime-switching approach. IMF Working Paper 03/32. Washington: International Monetary Fund
Angkinand A, Li J, Willett TD (2006) Measuring currency crises. Presented at the annual meetings of the Western Economic Association
Aziz J, Caramazza F, Salgano R (2000) Currency crises: in search of common elements. IMF Working Paper 00/67. Washington
Bordo M, Eichengreen B, Klingebiel D, Martinez-Peria MS (2001) Is the crisis problem growing more severe? Economic policy, CEPR & CES & MSH, vol. 16(32): 51–82, 04
Caramazza F, Ricci L, Salgado R (2000) Trade and financial contagion in currency crises. J Int Money Fin 23(1):51–70
Danielson de Haan L, Peng L, de Vries CG (2001) Using a bootstrap method to choose the sample fraction in tail index estimation. J Multivar Anal 76:226–248
De Haan L, Ferreira A (2006) Extreme value theory: an introduction. Springer, New York
Diebold F, Schuermann T, Stroughair J (1999) Pitfalls and opportunities in the use of extreme value theory and risk management. University of Pennsylvania, Mimeo
Eliasson AC, Kreuter C (2001) On currency crisis models: a continuous crisis definition. Deutsche Bank Research Quantitative Analysis unpublished working paper. http://www.economia.uniroma2.it/ceis/conferenz_convegni/banking2001/papers/mercoledi/Eliasson-Kreuter.pdf
Eichengreen B, Rose A, Wyplosz C (1995) Exchange market mayhem: the antecedents and aftermath of speculative attacks. Econ Policy 21(21):249–312
Eichengreen B, Rose A, Wyplosz C (1996) Contagious currency crises: first tests. Scand J Econ 98:463–484
Embrechts P, Kluppelberg C, Mikosch T (1997) Modeling extremal events for insurance and finance. Springer, Berlin
Enders W (2004) Applied econometric time series. Wiley, Hoboken, NJ
Flood RP, Marion NP (1999) Perspectives on the recent currency crisis literature. Int J Finance Econ 4(1):1–26
Frankel JA, Rose AK (1996) Currency crashes in emerging markets: an empirical treatment. J Int Econ 41(3–4):351–366
Glick R, Hutchison M (2001) Capital controls and exchange rate instability in developing economies. J Int Money and Fin 24(3):387–412
Girton L, Roper D (1977) A monetary model of exchange market pressure applied to postwar Canadian experience. Am Econ Rev 67:537–548
Haile FD, Pozo S (2008) Currency crisis contagion and the identification of transmission channels. Int Rev Econ Finance 17:572–588
Haile FD, Pozo S (2006) Exchange rate regimes and currency crises: an evaluation using extreme value theory. Rev Int Econ 14(4):554–570
Hill BM (1975) A simple general approach to inference about the tail of a distribution. Ann Stat 3(5): 1163–1174
Huisman R, Koedijk KG, Kool CJM, Palm F (2001) Tail-index estimates in small samples. J Bus Econ Stat 19(1):208–216
International Monetary Fund, International Financial Statistics: http://www.imfstatistics.org/imf/
Jansen DW, De Vries CG (1991) On the frequency of large stock returns: putting booms and busts into perspective. Rev Econ Stat 73:18–24
Kamin SB, Schindler JW, Samuel LS (2001) The contribution of domestic and external factors to emerging market devaluation crises: an early warning system approach. Board of Governors of the Federal Reserve System, International Finance Discussion Papers 711
Kaminsky G, Lizonado S, Reinhart CM (1998) Leading indicators of currency crises. Int Monet Fund Staff Pap 45:1–48
Kaminsky G, Reinhart CM (1998) Financial crises in Asia and Latin America: then and now. Am Econ Rev 88:444–448
Koedijk KG, Stork PA, de Vries CG (1992) Differences between foreign exchange rate regimes, the view from the tails. J Int Money Finance 11:462–473
Koedijk KG, Schafgans MMA, de Vries CG (1990) The tail index of exchange returns. J Int Econ 29:93–108
Krugman P (1979) A model of balance of payments crises. J Money Credit Bank 11:311–325
Lestano J, Jacobs Jan PAM (2007) Dating currency crises with ad hoc and extreme value-based thresholds: East Asia 1970–2002. Int J Finance Econ 12:371–388
Lee J, Strazicich M (2004) Minimum LM unit root test with one structural break. Working Paper, Department of Economics Applachian State University
Longin F, Solnik B (2001) Extreme correlation of international equity markets. J Finance 56(2):649–676
Martinez-Peria MS (2002) A regime switching approach to studying speculative attacks: a focus on European monetary system crises. Empir Econ 27:299–334
Pozo S, Amuedo-Dorantes C (2003) Statistical distributions and the identification of currency crises. J Int Money Finance 22(4):591–609
Pontines V, Siregar R (2008) Fundamental pitfalls of exchange market pressure-based approaches to identification of currency crises. Int Rev Econ Finance 17(3):345–365
Sachs JD, Tornel A, Velasco A (1996) Financial crises in emerging markets: the lessons from 1995. Brooking Pap Econ Act 1:147–215
Tudela M (2003) Explaining currency crises: a duration model approach. J Int Money Finance 23:799–816
Wagner N, Marsh TA (2005) Measuring tail thickness under GARCH and an application to extreme exchange rate changes. J Empir Finance 12:165–185
Wagner N, Marsh TA (2000) On adaptive tail index estimation for financial return models. Research program in finance working paper RPF-295 U.C. Berkeley: Haas School of Business
Willett T, Nitithanprapas E, Nitithanprapas I, Rongala R (2004) The Asian crises reexamined. Asian Econ Pap 3:32–87
Zhang Z (2001) Speculative attacks in the Asian crisis. IMF Working Paper 01/189. Washington
Acknowledgments
We thank two anonymous referees for helping us to significantly improve this current version of the paper. All remaining errors are our own.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
To determine the optimum cutoff value by Monte Carlo simulations, we adopt the simulation steps from Longin and Solnik (2001) and Haile and Pozo (2008) and run them with R software. These steps are as follows:
-
1.
\(S\) time series containing \(n\) observations of EMP are simulated. Each S is derived from a known Student’s \(t \)distribution with \(\alpha \) degrees of freedom, where \(\alpha \) ranges from 1 to \(K. \)The class of the Student’s \(t \)distributions chosen to consider different degrees of tail fatness. Since the tail index \(\gamma \) is inverse of \(\alpha \) (\(\gamma \)=1/\(\alpha )\), the lower the degree of freedom is, the fatter the distribution will be. In our simulation, \(\alpha \) is allowed to take values from 1 to 10 with increment of 0.1 and number of replications (\(S)\) equals 1,000.
-
2.
For different numbers of \(m\) of the extreme EMPs, a tail index \(\gamma _{s}^{h}(m\),\(\alpha )\) corresponding to the \(s-\)th replication from the Student’s \(t\) with \(\alpha \) degrees of freedom is estimated. Values of \(m\) can vary from %1 to %30 of \(n, \) where \(n\) is the sample size of the actual EMP data.
-
3.
For a particular Student’s \(t\) distribution with \(\alpha \) degrees of freedom and for each value of \(m\), MSE of the \(S \)tail index estimates, which is denoted by MSE (\(\gamma _{s}^{h}(m\),\(\alpha )_{s=1,2, }\)...,\(_{S})\), is computed. This computation repeatedly continues for different values of \(m\) and particular Student’s \(t\) with \(\alpha \) degrees of freedom. Then the optimal \(m\), denoted by \(m^{*}(\alpha )\), which minimizes MSE for the particular Student’s \(t \) distribution with \(\alpha \) degrees of freedom is selected.\(^{25}\) Optimum values of \(m \) for different Student’s \(t \) distributions are repeatedly selected. A total of \(K \) optimal values of \(m*, (m*(\alpha ))_{\alpha = 1,2, \ldots ,K}, \) are subsequently selected for \(K\) possible theoretical distributions.
-
4.
Using each of \(K\) optimum values of \(m\) that are obtained in last step, the Hill index, \(\gamma ^{h}(m*(\alpha )\), is estimated from actual EMP series. For all \(\alpha \) from 1 to 10, the tail indices, \(\gamma ^{h}\), are estimated from actual EMP series.
As the final step, we select one single number(say \(m**)\) from the \(K \)optimum tail indices, \(m*,\) for each EMP series such that the estimated tail index from the actual data (step 4) is statistically the closest to the corresponding tail index of the theoretical distribution. The main objective of the whole exercise is to determine the number of extreme observations for each EMP series. This value, \(m**\), that is corresponding to the optimum tail index, specifies the number of observations as the largest EMPs or episodes of currency crises.
Rights and permissions
About this article
Cite this article
Karimi, M., Voia, M. Identifying extreme values of exchange market pressure. Empir Econ 48, 1055–1078 (2015). https://doi.org/10.1007/s00181-014-0851-5
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00181-014-0851-5