Patterns of financial crisis: CHF/JPY 20072009 
Written by Forex Automaton  
Tuesday, 24 March 2009 14:56  
CHF/JPY dropped during the crisis, as did the interest rate differential between the currencies. The autocorrelation "pattern" of logarithmic returns resembles those of EUR/GBP, GBP/USD, EUR/JPY, and EUR/USD, with the hourscale "bipolar disorder", seen in the analysis as a significant negative correlation value at one hour lag, growing in prominence during the volatile phase of the crisis. Still, in CHF/JPY it falls short of the magnitude seen in EUR/CHF (ironically, one of the least volatile in forex), not to mention AUD/USD. In short, not just the increased volatility discussed in the media, but also the "bipolar disorder", a particular form of prehistory dependence, is seen as the signature pattern of financial panic.
I define the visible phase of the present financial crisis to begin on August 16, 2007, the day of Countrywide Financial nearbancruptcy event, followed by an extraoridinary halfpercent Fed discount rate cut next day. This study covers 82 weeks from August 2, 2007 through February 12, 2008. The choice of February 12 is motivated by purely technical considerations  there is no indication that the crisis ends there! The subrange of extreme volatility (as will be seen in Fig.4) can be roughly defined as the last 28 weeks of this range. In this study, I only look at trading activity taking place from 1am to 1pm New York time, since the experience shows it to be the richest in nontrivial correlations. Speaking of CHF/JPY trends, the advent of the high volatility regime coincides with CHF/JPY breaking out of the rangebound evolution in August 2008. Fig.2 shows the corresponding evolution of the autocorrelation of logarithmic returns in the vicinity of zero time lag. The correlation structure is represented as a triplet of correlation values: those at zero, one and two hour lags. The increased volatility shows up as the increase in the magnitude of all these values, with variance (a measure of volatility) being fairly well represented by the magnitude of the zero time lag value. Ideally one would like to be able to estimate significance of the observed correlation values. I do this by constructing many uncorrelated time series with volatility of the real one, running the analysis on them, and obtaining the mean and the variance of the correlation values at each time lag as a sample mean and variance  or mean and RMS. These are shown as the mean and the range of the red background sequence in Fig.3. The problem however is the increase in volatility from "phase 1" to "phase 2", which would make a simple approach based on stationary time series (that is, constructed on the basis of the same distribution) invalid. Since the question of potential changes brought about by the crisis is interesting in itself, we separate the "phase 1" and "phase 2" correlations, see Fig.3.
The pattern of "bipolar disorder" has been seen before in forex exchange rates with high interest rate differential at stake, and also in LIBOR time series analyses. The "bipolar disorder" feature, a tendency to form alternating rises and falls on nexthour time scale, more pronounced than in a fully unpredictable time series of the same volatility, shows up as negative deeps surrounding the zerotime lag peak. The difference between top and bottom panels of Fig.3 can be summed up as an increase in "bipolar disorder" during the volatile phase of the crisis. 

Last Updated ( Monday, 04 January 2010 12:38 ) 