Patterns of financial crisis: GBP/USD in 2007-2008.

User Rating: / 4
PoorBest 
Written by Forex Automaton   
Tuesday, 20 January 2009 12:42

The GBP/USD study is done in the same way as the previous EUR/JPY and EUR/USD studies: the time interval starting in August 2007 is split into "moderate" and high volatility parts, whose features are then compared. The rally of low-yielders (USD and JPY) that characterized the high volatility phase seems to distort the autocorrelation structure in a very particular way, and this influence is found to be qualitatively the same for the exchange rates looked at so far.

Evolution of GBP/USD exchange rate during the financial crisis, hour.

Fig.1:GBP/USD during the financial crisis, hour time scale. Time axis is labeled in MM-YY format and spans the interval from August 2, 2007 through December 31, 2008.

I define the visible phase of the present financial crisis to begin on August 16, 2007, the day of Countrywide Financial near-bankruptcy event, followed by an extraordinary half-percent Fed discount rate cut next day. This study covers 74 weeks from August 2, 2007 through December 31, 2008. The sub-range of extreme volatility (as will be seen in Fig.4) can be roughly defined as the last 18 weeks of this 74-week range. Thus, I divide the time interval under study in two unequal parts, those of (relatively speaking) low and high volatility. The low volatility phase is from August 2, 2007 through August 27, 2008. The high volatility phase is from August 28 through the end of 2008. 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 non-trivial correlations.

GBP/USD volatility change during the financial crisis, hour.

Fig.2:The histogram of logarithmic returns in GBP/USD on the hour time scale demonstrates volatility change in the course of the financial crisis.

While the change in the volatility between the beginning of the crisis and its "phase of impact" is undeniable, single-point distributions like Fig.2 do not tell the whole story. These distributions could belong to a random walk or to a more complex pattern where prehistory matters.

Autocorrelation of logarithmic returns in GBP/USD,  European trading hours, hour scale, from August 2, 2007 through August 27, 2008. Autocorrelation of logarithmic returns in GBP/USD,  European trading hours, hour scale, from August 28, 2008 through December 31, 2008.

Fig.3: Autocorrelation of logarithmic returns in GBP/USD for the European (Eurasian) trading shown against the backdrop of statistical noise (red). Top panel: the measurement time range is for the relatively low volatility phase of the crisis, from August 2, 2007 through August 27, 2008. Bottom panel: same for the high volatility phase, from August 28, 2008 through the end of 2008. The noise is obtained from martingale simulations based on the recorded volatilities of GBP/USD in the trading hours under study for the period. The noise is presented as mean plus-minus 1 RMS, where RMS characterizes the distribution of the correlation value obtained for each particular bin by analyzing 20 independent simulated uncorrelated time series of the same average volatility. From top to bottom, the shape of the autocorrelation in the vicinity of the zero-time lag peak undergoes a remarkable transformation.This is to be compared with the analogous figures for the EUR/JPY and EUR/USD autocorrelations.

In addition to the fact that the volatility got higher in the impact phase of the crisis (making the Fig.2 distribution wider and Fig.3 -- higher), a correlation pattern new to GBP/USD -- the pattern of "bipolar disorder" -- appears in the bottom panel of Fig.3. This pattern 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 quickly alternating rises and falls on next-hour time scale, more pronounced than in a fully unpredictable time series of the same volatility, shows up as negative deeps surrounding the zero-time lag peak. These have significance of two standard deviations in the time-integrated correlation of Fig.3, bottom panel. The practical implication of a feature like this is simple: since any movement is likely to be followed by the movement in the opposite direction, it is hard to make money on trend following. Rephrasing the Wall Street adage, bulls get slaughtered, bears get slaughtered, pigs get slaughtered, contrarians (and brokers) make money. In fact, both top and bottom panels of Fig.4 look similar to their counterparts from EUR/JPY and EUR/USD. The "bipolar disorder" effect in the bottom panel is common to EUR/JPY, EUR/USD, and GBP/USD (the case under study) and in all three cases it is similarly marginally statistically significant.

Evolution of GBP/USD autocorrelation peak structure during the financial crisis, hour.

Fig.4: Evolution of GBP/USD autocorrelation peak structure during the financial crisis, hour time scale. Time bin is two weeks wide. The peak structure is represented by three correlation values: the one for the zero lag (essentially a volatility measure) downscaled by 10 for easier visual comparison, the one at one hour lag (just discussed) and the one at two hour lag. Time axis is labeled in MM-YY format and spans the interval from August 2, 2007 through December 31, 2008. Only trading hours from 1am to 1pm New York time (European trading hours), usually rich in non-trivial correlations, are included.

The time period of dramatically higher volatility covers the last 9 bins in Fig.4, which contain 18 weeks. The volatility can be judged by the magnitude of the zero time-lag peak, since it's approximately equals variance for a time series with such a negligible mean. The magnitude of the negative one-hour lagged correlation, a convenient measure of the strength of the bipolar disorder syndrome, is seen to become truly monstrous in the last couple of weeks of 2008 (see historical GBP/USD chart) -- predictably, that was the time of a US Dollar rally.

The data used are from the period 2002-08-02 00:00:00 to 2009-01-01 00:00:00, New York time.

Bookmark with:

Deli.cio.us    Digg    reddit    Facebook    StumbleUpon    Newsvine
Last Updated ( Monday, 04 January 2010 12:41 )