AUD/USD "bipolar disorder" history

User Rating: / 7
PoorBest 
Written by Forex Automaton   
Wednesday, 29 October 2008 17:36

This is a follow-up note to AUD/USD: predictability overview. AUD/USD is one of the currency pairs with what I call hour-scale "bipolar disorder" predictability 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 same volatility. Any statistical test of market inefficiency requires a finite time of observation, and the original article covered the time interval from August 20, 2002 to February 1st, 2008, with no attention paid to the time evolution of the picture of predictable patterns in this currency pair. This article extends the historical coverage into the first three quarters of 2008, and focuses on the time evolution of the pattern, characterized by the 1-hour time lag autocorrelation value. Naturally, we want to know how robust this feature has been, whether it is alive at present, and what are its future prospects.

AUD/USD historical chart 2002-2008

Fig.1: Bar chart of AUD/USD history, 2002-2008. The time axis is labeled in MM-YY format.

AUD/USD autocorrelation of logarithmic returns in the vicinity of zero time lag, hour scale, separating time zones

Fig.2:Autocorrelations of logarithmic returns in AUD/USD, over a range of time lags, computed for the time period October 1, 2002-October 1, 2008. Time scale of the analysis is hour.

For the sake of completeness, Fig.1 presents the history of AUD/USD for the period under study. Fig.2 portrays the subject of this note -- the "bipolar disorder", using the psychiatric analogy ascribed to Benjamin Graham ("Intelligent Investor"). Justifying the psychiatric analogy, the large negative autocorrelation signal in the bin next to zero indicates rapid (next hour) changes in the mood of the market, a price action followed by a next hour correction. Technically, these represent a profit opportunity to a speculator able to stay cool and take advantage of the market's excesses, entering after the "action" and taking profit on the predictable "reaction". What is not clear from Fig.2 is whether these events happen systematically.

Time history of the one hour time lag value of the autocorrelation of logarithmic returns, AUD/USD

Fig.3: Time history of the one hour time lag value of the autocorrelation of logarithmic returns in AUD/USD. The time axis is labeled in MM-YY format. All time zones.

To trace the time evolution of the effect, Fig.3 plots the correlation value at the -1 hour time lag over time. Plotted as red background is the noise estimate, obtained from martingale simulations based on the historical volatilities of AUD/USD for the period under study. The noise is presented as mean plus-minus 1 RMS, where the RMS characterizes distribution of the correlation value obtained for this particular time lag bin by analyzing 20 independent simulated pairs of uncorrelated time series. The RMS is a measure of accuracy in the determination of the correlation values, an uncertainty dependent on the amount of data and the time scale. This shows that the feature is statistically significant.

Time history of variance, or the peak value of the autocorrelation of logarithmic returns, AUD/USD

Fig.4: Time history of variance, or the peak value of the autocorrelation of logarithmic returns in AUD/USD. The time axis is labeled in MM-YY format. All time zones.

Fig.4, time evolution of variance, demonstrates that the effect in Fig.3 is even more significant than it looks -- as the next figure will show, the average volatility for the period is strongly influenced by the most recent data, 3rd quarter of 2008, and is therefore an overestimate for the overall period 2002-2008). There is another important message: volatility varies strongly and apparently, the time variation of the strength of the effect is entirely due to a change in the volatility with time. Checking this is best done by representing the effect not as an absolute correlation, but as a Pearson correlation coefficient -- covariance normalized to the variance.

One hour time lag magnitude of the AUD/USD autocorrelation of logarithmic returns, 2002-2008, shown as Pearson correlation coefficent, all time zones One hour time lag magnitude of the AUD/USD autocorrelation of logarithmic returns, 2002-2008, shown as Pearson correlation coefficentm, European trading

Fig.5:One hour time lag magnitude of the Pearson-normalized autocorrelation of logarithmic returns in AUD/USD, plotted as a function of time, October 1, 2002 through October 1, 2008. Time axis is labeled in MM-YY format. Top: all time zones, bottom: European trading (1am through 1pm New York time). The number of bins (24) equals the number of quarters in the six-year period. The noise (red in the figure) is obtained from martingale simulations based on the historical volatilities of AUD/JPY for the period under study. The noise is presented as mean plus-minus 1 RMS, where the RMS characterizes distribution of the correlation value obtained for this particular time lag bin by analyzing 20 independent simulated pairs of uncorrelated time series. The RMS is a measure of accuracy in the determination of the correlation values, an uncertainty dependent on the amount of data and the time scale.

Presenting the normalized magnitude of the one hour time lag autocorrelation as a function of time, Fig.5 has a few important messages:

  • We are not dealing with just a single once-in-a-lifetime event. The inefficiency in question ("bipolar disorder") is historically continuous.
  • The heyday of the feature seems to be the period of late 2003 -- early 2005, as can be judged from the magnitude of the correlation. (Same was said of the similar feature in AUD/JPY).
  • The feature was still there in early 2008.

 

History of s/n-o/n LIBOR interest rate differential between AUD and USD, 2002-2008

Fig.6:History of the s/n-o/n LIBOR rate differential between AUD and USD.

Comparison of the time histories of the LIBOR rate differential, Fig.6, and variance, a measure of volatility, Fig.4, reveals that the two closely follow one another -- interest rate differential creates volatility. It would be harder to say the same about the time-lag 1 effect -- at least with the European trading data from Fig.5 -- the effect appears and disappears, but at least one thing is certain: there is either the negative predictive correlation or no predictive correlation -- a positive correlation such as would be required to justify momentum-trading or trend following in AUD/JPY, is not there.

Bookmark with:

Deli.cio.us    Digg    reddit    Facebook    StumbleUpon    Newsvine
Last Updated ( Monday, 14 September 2009 16:52 )