Scribd Logo
Upload

Welcome to Scribd!

  • Foreign Exchange System PDF

    Uploaded by

    Mouktar Abdi Djama
    30 views10 pages

    Original Title

    foreign exchange system.pdf

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    Download as pdf
    30 views10 pages

    Foreign Exchange System PDF

    Uploaded by

    Mouktar Abdi Djama

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    Download as pdf
    of 10
    E. van der Merwe Smit, TW. Pahn South African Foreign Exchange Risk under Managed Floating: Distributional Aspects ABSTRACT ‘The paper examines exchange rate risk, defined as the varia: biltyn weekly log-ratios ofthe Rand-Dollar exchange rate dur- ing the decade of managed floating. Distributional tests lead tothe rejection of the Gaussian model often used to evaluate ‘exchange rate risk. Estimation shows that the leptokurtic ‘character of the empirical distributions is better characterised by the class of non-normal stable Paretian distributions, 1. INTRODUCTION. ‘This paper examines exchange rate risk, defined as the varia: bility in the log-ratios of the land-Dollar exchange rate, dur- ing the regimo of managed floating. The scope of the ernpir ‘al research is limited to the nominal commercial Rand vs, the US. Dollar spot exchange rate, as wel as othe correspond- ing one, three, six and twelve month forward rates. As for the ‘spot rate, the period under consideration starts with the in- ‘reduction of the managed floading regime in January 1979, whereas the examination of the forward rates begins in Sep- tember 1983, marking the point when the South African Reserve Bank terminated the system of administored forward Contracts in favour ofa swap system, Both sample periods end in March 1990. The purpose of this study isto investigate whether the risk as ‘measured by the variability of the Rand-Dollar exchange rate was subject 1 notable changes given the external pressures. and changes in domestic monetary policy during this period The sample data are thus divided into various sub-periods, coinciding with major events during the past decade. Special Consideration is given to the unrest period between August 1984 and October 1986 Although itis commonly assumed in foreign exchange re- search thal the Gaussian or normal model accounts or sue- cessive diferences in exchange rates, the distributional tests of ths study load to the rejection of the normal modo in the case ofthe Rand due the highly leptokutic character ofthe empirical distributions. Atornatve statistical distbutions have tobe employed to provide more adequate models ofthe be- haviour of exchange rate changes. The assumption of areal ‘World situation similar to changos of stock-market prices (Man- dolorot, 1963), leads to the alternative hypothesis of an un- deriving form of non-normal stable Paratian distributions, Un- tke the normal model which is completely described by is mo- ments, the non-normal stable Paretian distributions are desert bed by their characteristic exponents and their soale and location parameters. The formers elated to the tal area probe- bilty ofthe density function. An important characteristic ofthe stable Paretian family of distributions isthe infinite variance “Thus the variance isan inadequate measure of variability both ‘theoretically and empirically. Consequently the sample stan- 1, /. In this case the location parameter, resembles the mean (Fama and Roll, 1968:826)). The Cauchy distribution with « = 1, s another member of the class of sym- metric stable distributions. Ithe characteristic exponent approaches a value smaller than 2, the variance and other higher moments cease to exist. The far reaching implications of this characteristic are explicitly ex- plained by Fama (1965b:94-95): “The statistical implications, of the Mandelbrot hypothesis follow mostly from the absence of variance forthe stable Paretian distributions with a charac- teristic exponent less than 2. In practical terms "infnite”’means ‘that the sample variance and standard deviation of a stable Paretian process with « < 2 will show extremely erratic be- haviour even for large samples. That is, for larger and larger, ‘sample sizes the variability of the sample variance and stan- dard deviation willnat tend to dampen nearly as much as would be expected with a Guassian process. Because of their ex: tremely erratic behaviour, the sample variance and standard deviation are not meaningful measures of the variability in herent in a stable Paretian process with « < 2. Investment Analysts Journal ~ Summer 198211988 8 ‘South African Foreign Exchange Risk under Managed Floating: 63 Goodness-of-fit tests ‘The Chi-square statistic of goodness-ofit provides a means to assess whether the sample distributions are likely to be drawn fom the population of symmetric stable distributions with @ given characteristic exponent. Here the Chi-square Siatistic is used to examine which value (1 < «= 2) best de- ‘scribes the empirical distributions. Thus the statistic is com- puted forall spot and forward rate distributions for characteris- tic exponent values from 1,1 to 2,0, The results of the tests are presented in Table 2. The testis based on findings of Fama ‘and Roll (1968:817-836) who published fraciles of standardised ‘symmetric stable distributions, generated by means of a Monte- Carlo study. TABLE 2 CHI-SQUARE TESTS SADREAL StS B me Fe 1335515215 14808 5g 762 7.279" 8233771" 97 10990 12599 12,192 2306 © 25.808 30,099 4139 41.009 49488 52657 48.849 50855 58674 5748363413 6169 65797 R994 12657 73,703 78,180 4 tat 4.22 RNDDPI NDDPI v70 — 4785 WBS ~ 5180 Nest Na248 apne 10,878 16,82" 70881 alpha 9082" 17559 659" alpha = 13 11,180, 2278 alpha = 1414874 27383 alpha = 15 28.604 5.412 abba = 18 34001 7,324 apa = 17 53.03 apna = 18 §2374 ‘apa = 1983785 ‘pha = 20 Estate EALPL FB Fé ‘yas - 25785 alpha = 41 175600 170.200 178.000 181,600 178,500 alpha = 42 163300 169,300" 161,600 180200 174,600, abha = 13° 172000 188000 171.900" 173.700 173200 ‘pha = 14 159200" 166400 172400 164.100" 170,800" aipha = 15 163,100 166800 17,300 172500 171,700 ‘pha = 18 179900 176900 188.200 © 173.900 181200 ‘pha = 17 189,100 186700 203.800 183,700 186300 alpha = 18 212700 210,700 232.100 226,100 210,700 alpha = 19 241900 297,500 279800 270,100 261,100 ‘aha = 20 268.00 266.00 286,600 279300 251,000 Estimate 120408 go 130, REALPIL Spot Fr FoF ‘fae — 291980 Neada alpha 9e75 1158314852 15.467 18,350, alpha Sooo Sarr" aia" 10057" gas9 ‘pna= 13 6935918 Bast 10.754 7809" 49790 784412025 10959 12.148 fipha = 15 16020 17.148 19.648 zest 28582 alpha = 16 25057 24505 2aait 26782567 alpha = 17 92107 91.248 90,193 32.270 83.664, alpha = 18 S088 96164 95.221 38598 95.590, aipha = 19 36820 34.483 4091839689 40.508 aha = 20 46451 358245958 W957 45918, Estimate 428128128 ‘START BEGIN 79 — 2088 ‘ies ~ 26784 Newe A Fa apha= 125877 12000 10,889 apha= 12 2582912417 10,839" apha= 13 21004 13687 13250 alpha = 14 14083 15,750 aipha = 15 10393 12.000 alpha = 18 10333 12000 pha = 17 4088" 13.687 alpha = 12 4917 13667 alpha = 19 4917 13687 ‘alpha = 20 4917 13867 10.750 Estimate 1a 50 ist UNREST A 8 Fe 2a — Taos Neti6 apha= 1115550 17601 1572414608 20582 1290" 15290" 14086" 19000" 18828 16,759 2248 19088 18052 19948 20,698, 20704 22017 21,808 24890 mo 23397252071 27278 © 78.489 25.588 40828 27am 30466 © 35.121 58.000 1500 42.780 47.448 65293 431054431 82.988 78781 528125 tak Ft 5 FoF 28.944 igor Dt,94e 1527816833 10333 9833 128. 7278 6058 480 3885388 att 4556" 5000 5687 5000 5,687 17482 ‘The sample values are first transformed into standardised - values. The random sample, 1. the log-ratios, x, | = 1, N, where i equals the weekly observation for each spot and for ward rate, were standardised by subtracting an estimate of the mean X and dividing by the estimate of the scale parameter © (also dispersion parameter) to obtain the statistic (x, — X)/c Fama and Roll (1968:832) found that in cases where ois, ‘unknown, 0:50 truncated means generally perform as "best estimators" over the entire range of 1 < a < 2.Asinthe proce- dure used by Westerfield (1977:186 footnote 7), the stan- dardised empirical samples were then ranked in ascending ‘order and divided into deciles. To examine the extreme tail areas more closely, the last decile in both tals was again divi €1 in half (only exception: fle BEGIN ~ the small number of ‘observations made a division into only ten class intervals pos- sible). The number of observations in each of the twelve (ten) cells were counted and the Chi-square computed for nine Investment Analysts Journal ~ Summer 1992/1993 ‘South African Foreign Exchange Risk under Managed Floating: Distributional Aspects degrees of freacom (the location and scale parameters had 10 be estimated in order to transform the log-ratios into stan- dardised t-values, thus two degrees of freedom were lost). The period of unrest displays observations that are more than five Standard deviations from the mean. Under the Guassian hypothesis such observations should only occur once every "7000 years (Fama, 1965b:50). The number of observations in the tails are much larger than expectod under the normal model; thus the probabil of larger price changes is much higher In aocordance with the findings of Westerfield (1977:189) the sample distributions are more peaked in the centre, whereas the remaining areas between the centre and the tails display fewer observations than expected. ‘The examination of the results in Table 2 shows that the nor- ‘mal model provides an inadequate description of the epi cal distributions when compared to other symmetric stable dis- tributions, The Chi-square value when x = 2is the largest in the majority of cases and decreases gradually as the value ‘of the charactoristic exponent Is reduced to an average level ‘of 13 (which is exactly the samo behaviour as observed by ‘Westerfield, 1077-188). The Chi-square value inereases again as the characteristic exponent further approaches the Cauchy distribution ‘Testing the null hypothesis thatthe empirical distributions are drawn from the population of symmetric stable distrioutions With a given characteristic exponent 1 < a <2, when the value of the Chi-square statistic is at a minimum, ieads to accep- tance at a 0025 level of significance in 26 out of 33 or seventy. tight percent ofthe cases. tis intoresting to note that the null- hypothesis had tobe rejected exclusively for the Pre-Rubicon ‘samples. Nevertheless, the null hypothesis is accepted in more than seventy percent of the cases, as in Westerfield’s study (1977:189), which suggests a high degree of appraximation. In Rana's examination of the behaviour of the exchange rate ‘of eight Asian countries (1980:7) the null hypothesis was ac- Copied in all but ono of the cases. The Chi-square statistic of goodness-of-fit allows the conclu- sion that the empirical samples are better described by mem- bers of the class of symmetric non-normal stable distributions with 1

    You might also like